matrixgrad.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Learning the basics of neural nets by implementing basic algos from scratch. This notebook is used to test and demo matrixgrad.py. It is is based on Andrej Karpathy's https://github.com/karpathy/micrograd applied to 2d numpy arrays; so kind of a blend of his micrograd and backprop ninja online classes in https://karpathy.ai/zero-to-hero.html"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, let's test and look at the autograd layer in matrixgrad.py. The code below creates a graph function to visualize a DAG of Matrix classes and uses that to visualize the graph created by performing operations on Matrices"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/svg+xml": [
       "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n",
       "<!DOCTYPE svg PUBLIC \"-//W3C//DTD SVG 1.1//EN\"\n",
       " \"http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd\">\n",
       "<!-- Generated by graphviz version 2.50.0 (0)\n",
       " -->\n",
       "<!-- Pages: 1 -->\n",
       "<svg width=\"2816pt\" height=\"176pt\"\n",
       " viewBox=\"0.00 0.00 2815.89 175.50\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n",
       "<g id=\"graph0\" class=\"graph\" transform=\"scale(1 1) rotate(0) translate(4 171.5)\">\n",
       "<polygon fill=\"white\" stroke=\"transparent\" points=\"-4,4 -4,-171.5 2811.89,-171.5 2811.89,4 -4,4\"/>\n",
       "<!-- 1745400059216 -->\n",
       "<g id=\"node1\" class=\"node\">\n",
       "<title>1745400059216</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"2513.89\" cy=\"-149.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"2513.89\" y=\"-145.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">0.5</text>\n",
       "</g>\n",
       "<!-- 1745401954656* -->\n",
       "<g id=\"node18\" class=\"node\">\n",
       "<title>1745401954656*</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"2632.89\" cy=\"-119.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"2632.89\" y=\"-115.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">*</text>\n",
       "</g>\n",
       "<!-- 1745400059216&#45;&gt;1745401954656* -->\n",
       "<g id=\"edge14\" class=\"edge\">\n",
       "<title>1745400059216&#45;&gt;1745401954656*</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2539.41,-143.21C2556.26,-138.89 2578.86,-133.1 2597.51,-128.32\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"2598.57,-131.65 2607.39,-125.78 2596.84,-124.87 2598.57,-131.65\"/>\n",
       "</g>\n",
       "<!-- 1744877508096 -->\n",
       "<g id=\"node2\" class=\"node\">\n",
       "<title>1744877508096</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"290,-65.5 290,-111.5 463,-111.5 463,-65.5 290,-65.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"376.5\" y=\"-96.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[ 1. 4.] [ 2. 5.] [&#45;3. 6.]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"290,-88.5 463,-88.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"376.5\" y=\"-73.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[0. 0.] [0. 0.] [0. 0.]]</text>\n",
       "</g>\n",
       "<!-- 1745401739824@ -->\n",
       "<g id=\"node5\" class=\"node\">\n",
       "<title>1745401739824@</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"526\" cy=\"-55.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"526\" y=\"-51.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">@</text>\n",
       "</g>\n",
       "<!-- 1744877508096&#45;&gt;1745401739824@ -->\n",
       "<g id=\"edge20\" class=\"edge\">\n",
       "<title>1744877508096&#45;&gt;1745401739824@</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M463.3,-69.3C472.75,-67.19 481.92,-65.14 490.19,-63.29\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"491.12,-66.67 500.11,-61.07 489.59,-59.84 491.12,-66.67\"/>\n",
       "</g>\n",
       "<!-- 1744877508096T -->\n",
       "<g id=\"node3\" class=\"node\">\n",
       "<title>1744877508096T</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"227\" cy=\"-88.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"227\" y=\"-84.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">T</text>\n",
       "</g>\n",
       "<!-- 1744877508096T&#45;&gt;1744877508096 -->\n",
       "<g id=\"edge1\" class=\"edge\">\n",
       "<title>1744877508096T&#45;&gt;1744877508096</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M254.03,-88.5C261.62,-88.5 270.4,-88.5 279.69,-88.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"279.93,-92 289.93,-88.5 279.93,-85 279.93,-92\"/>\n",
       "</g>\n",
       "<!-- 1745401739824 -->\n",
       "<g id=\"node4\" class=\"node\">\n",
       "<title>1745401739824</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"589,-32.5 589,-78.5 686,-78.5 686,-32.5 589,-32.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"637.5\" y=\"-63.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[&#45;4. 32.]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"589,-55.5 686,-55.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"637.5\" y=\"-40.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[0. 0.]]</text>\n",
       "</g>\n",
       "<!-- 1745401740016ReLU -->\n",
       "<g id=\"node15\" class=\"node\">\n",
       "<title>1745401740016ReLU</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"754.5\" cy=\"-55.5\" rx=\"32.49\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"754.5\" y=\"-51.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">ReLU</text>\n",
       "</g>\n",
       "<!-- 1745401739824&#45;&gt;1745401740016ReLU -->\n",
       "<g id=\"edge19\" class=\"edge\">\n",
       "<title>1745401739824&#45;&gt;1745401740016ReLU</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M686.27,-55.5C694.69,-55.5 703.39,-55.5 711.63,-55.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"711.77,-59 721.77,-55.5 711.77,-52 711.77,-59\"/>\n",
       "</g>\n",
       "<!-- 1745401739824@&#45;&gt;1745401739824 -->\n",
       "<g id=\"edge2\" class=\"edge\">\n",
       "<title>1745401739824@&#45;&gt;1745401739824</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M553.23,-55.5C560.95,-55.5 569.72,-55.5 578.61,-55.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"578.64,-59 588.64,-55.5 578.64,-52 578.64,-59\"/>\n",
       "</g>\n",
       "<!-- 1745401954416 -->\n",
       "<g id=\"node6\" class=\"node\">\n",
       "<title>1745401954416</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"2457.89,-66.5 2457.89,-112.5 2569.89,-112.5 2569.89,-66.5 2457.89,-66.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"2513.89\" y=\"-97.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[86.44464]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"2457.89,-89.5 2569.89,-89.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"2513.89\" y=\"-74.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[0.5]]</text>\n",
       "</g>\n",
       "<!-- 1745401954416&#45;&gt;1745401954656* -->\n",
       "<g id=\"edge24\" class=\"edge\">\n",
       "<title>1745401954416&#45;&gt;1745401954656*</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2569.96,-103.62C2579.42,-106.05 2589,-108.5 2597.7,-110.73\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"2596.9,-114.14 2607.46,-113.24 2598.64,-107.36 2596.9,-114.14\"/>\n",
       "</g>\n",
       "<!-- 1745401954416sum -->\n",
       "<g id=\"node7\" class=\"node\">\n",
       "<title>1745401954416sum</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"2394.89\" cy=\"-89.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"2394.89\" y=\"-85.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">sum</text>\n",
       "</g>\n",
       "<!-- 1745401954416sum&#45;&gt;1745401954416 -->\n",
       "<g id=\"edge3\" class=\"edge\">\n",
       "<title>1745401954416sum&#45;&gt;1745401954416</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2422.14,-89.5C2429.78,-89.5 2438.48,-89.5 2447.41,-89.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"2447.53,-93 2457.53,-89.5 2447.53,-86 2447.53,-93\"/>\n",
       "</g>\n",
       "<!-- 1745401740928 -->\n",
       "<g id=\"node8\" class=\"node\">\n",
       "<title>1745401740928</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"1931.89,-66.5 1931.89,-112.5 2062.89,-112.5 2062.89,-66.5 1931.89,-66.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"1997.39\" y=\"-97.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[6.119203 7. ]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"1931.89,-89.5 2062.89,-89.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"1997.39\" y=\"-74.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[6.119203 7. ]]</text>\n",
       "</g>\n",
       "<!-- 1745401741168**2 -->\n",
       "<g id=\"node20\" class=\"node\">\n",
       "<title>1745401741168**2</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"2125.89\" cy=\"-89.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"2125.89\" y=\"-85.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">**2</text>\n",
       "</g>\n",
       "<!-- 1745401740928&#45;&gt;1745401741168**2 -->\n",
       "<g id=\"edge25\" class=\"edge\">\n",
       "<title>1745401740928&#45;&gt;1745401741168**2</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2063.12,-89.5C2071.91,-89.5 2080.66,-89.5 2088.7,-89.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"2088.79,-93 2098.79,-89.5 2088.79,-86 2088.79,-93\"/>\n",
       "</g>\n",
       "<!-- 1745401740928+ -->\n",
       "<g id=\"node9\" class=\"node\">\n",
       "<title>1745401740928+</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"1868.89\" cy=\"-89.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"1868.89\" y=\"-85.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">+</text>\n",
       "</g>\n",
       "<!-- 1745401740928+&#45;&gt;1745401740928 -->\n",
       "<g id=\"edge4\" class=\"edge\">\n",
       "<title>1745401740928+&#45;&gt;1745401740928</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1896.1,-89.5C1903.8,-89.5 1912.64,-89.5 1921.81,-89.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1921.85,-93 1931.85,-89.5 1921.85,-86 1921.85,-93\"/>\n",
       "</g>\n",
       "<!-- 1745401740448 -->\n",
       "<g id=\"node10\" class=\"node\">\n",
       "<title>1745401740448</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"1400.89,-64.5 1400.89,-110.5 1540.89,-110.5 1540.89,-64.5 1400.89,-64.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"1470.89\" y=\"-95.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[0.880797 1. ]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"1400.89,-87.5 1540.89,-87.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"1470.89\" y=\"-72.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;6.119203 &#45;7. ]]</text>\n",
       "</g>\n",
       "<!-- 1745401740688* -->\n",
       "<g id=\"node23\" class=\"node\">\n",
       "<title>1745401740688*</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"1603.89\" cy=\"-57.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"1603.89\" y=\"-53.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">*</text>\n",
       "</g>\n",
       "<!-- 1745401740448&#45;&gt;1745401740688* -->\n",
       "<g id=\"edge11\" class=\"edge\">\n",
       "<title>1745401740448&#45;&gt;1745401740688*</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1541.22,-71.62C1550.65,-69.46 1559.97,-67.33 1568.39,-65.4\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1569.21,-68.8 1578.18,-63.16 1567.65,-61.98 1569.21,-68.8\"/>\n",
       "</g>\n",
       "<!-- 1745401740448Sigmoid -->\n",
       "<g id=\"node11\" class=\"node\">\n",
       "<title>1745401740448Sigmoid</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"1323.94\" cy=\"-87.5\" rx=\"40.89\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"1323.94\" y=\"-83.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">Sigmoid</text>\n",
       "</g>\n",
       "<!-- 1745401740448Sigmoid&#45;&gt;1745401740448 -->\n",
       "<g id=\"edge5\" class=\"edge\">\n",
       "<title>1745401740448Sigmoid&#45;&gt;1745401740448</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1364.93,-87.5C1373.03,-87.5 1381.81,-87.5 1390.72,-87.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1390.82,-91 1400.82,-87.5 1390.82,-84 1390.82,-91\"/>\n",
       "</g>\n",
       "<!-- 1744875846336 -->\n",
       "<g id=\"node12\" class=\"node\">\n",
       "<title>1744875846336</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"822.99,-97.5 822.99,-143.5 971.99,-143.5 971.99,-97.5 822.99,-97.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"897.49\" y=\"-128.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[2. 3.]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"822.99,-120.5 971.99,-120.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"897.49\" y=\"-105.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.64247733 0. ]]</text>\n",
       "</g>\n",
       "<!-- 1745401740256+ -->\n",
       "<g id=\"node26\" class=\"node\">\n",
       "<title>1745401740256+</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"1034.99\" cy=\"-87.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"1034.99\" y=\"-83.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">+</text>\n",
       "</g>\n",
       "<!-- 1744875846336&#45;&gt;1745401740256+ -->\n",
       "<g id=\"edge15\" class=\"edge\">\n",
       "<title>1744875846336&#45;&gt;1745401740256+</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M972.19,-102.55C981.72,-100.23 991.1,-97.95 999.55,-95.89\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1000.47,-99.27 1009.35,-93.5 998.81,-92.47 1000.47,-99.27\"/>\n",
       "</g>\n",
       "<!-- 1745401739488 -->\n",
       "<g id=\"node13\" class=\"node\">\n",
       "<title>1745401739488</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"1670.89,-99.5 1670.89,-145.5 1801.89,-145.5 1801.89,-99.5 1670.89,-99.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"1736.39\" y=\"-130.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[7. 8.]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"1670.89,-122.5 1801.89,-122.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"1736.39\" y=\"-107.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[6.119203 7. ]]</text>\n",
       "</g>\n",
       "<!-- 1745401739488&#45;&gt;1745401740928+ -->\n",
       "<g id=\"edge13\" class=\"edge\">\n",
       "<title>1745401739488&#45;&gt;1745401740928+</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1802.25,-106.1C1813.06,-103.36 1823.88,-100.63 1833.53,-98.19\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1834.64,-101.52 1843.48,-95.67 1832.93,-94.73 1834.64,-101.52\"/>\n",
       "</g>\n",
       "<!-- 1745401740016 -->\n",
       "<g id=\"node14\" class=\"node\">\n",
       "<title>1745401740016</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"822.99,-32.5 822.99,-78.5 971.99,-78.5 971.99,-32.5 822.99,-32.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"897.49\" y=\"-63.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[ 0. 32.]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"822.99,-55.5 971.99,-55.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"897.49\" y=\"-40.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.64247733 0. ]]</text>\n",
       "</g>\n",
       "<!-- 1745401740016&#45;&gt;1745401740256+ -->\n",
       "<g id=\"edge21\" class=\"edge\">\n",
       "<title>1745401740016&#45;&gt;1745401740256+</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M972.19,-72.9C981.72,-75.16 991.1,-77.37 999.55,-79.37\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"998.82,-82.79 1009.35,-81.68 1000.43,-75.98 998.82,-82.79\"/>\n",
       "</g>\n",
       "<!-- 1745401740016ReLU&#45;&gt;1745401740016 -->\n",
       "<g id=\"edge6\" class=\"edge\">\n",
       "<title>1745401740016ReLU&#45;&gt;1745401740016</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M787.14,-55.5C794.93,-55.5 803.63,-55.5 812.64,-55.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"812.91,-59 822.91,-55.5 812.91,-52 812.91,-59\"/>\n",
       "</g>\n",
       "<!-- 1744875845952 -->\n",
       "<g id=\"node16\" class=\"node\">\n",
       "<title>1744875845952</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"0,-65.5 0,-111.5 164,-111.5 164,-65.5 0,-65.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"82\" y=\"-96.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[ 1. 2. &#45;3.] [ 4. 5. 6.]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"0,-88.5 164,-88.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"82\" y=\"-73.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[0. 0. 0.] [0. 0. 0.]]</text>\n",
       "</g>\n",
       "<!-- 1744875845952&#45;&gt;1744877508096T -->\n",
       "<g id=\"edge17\" class=\"edge\">\n",
       "<title>1744875845952&#45;&gt;1744877508096T</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M164.11,-88.5C172.95,-88.5 181.59,-88.5 189.48,-88.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"189.72,-92 199.72,-88.5 189.72,-85 189.72,-92\"/>\n",
       "</g>\n",
       "<!-- 1745401954656 -->\n",
       "<g id=\"node17\" class=\"node\">\n",
       "<title>1745401954656</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"2695.89,-96.5 2695.89,-142.5 2807.89,-142.5 2807.89,-96.5 2695.89,-96.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"2751.89\" y=\"-127.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[43.22232]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"2695.89,-119.5 2807.89,-119.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"2751.89\" y=\"-104.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad 1</text>\n",
       "</g>\n",
       "<!-- 1745401954656*&#45;&gt;1745401954656 -->\n",
       "<g id=\"edge7\" class=\"edge\">\n",
       "<title>1745401954656*&#45;&gt;1745401954656</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2660.14,-119.5C2667.78,-119.5 2676.48,-119.5 2685.41,-119.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"2685.53,-123 2695.53,-119.5 2685.53,-116 2685.53,-123\"/>\n",
       "</g>\n",
       "<!-- 1745401741168 -->\n",
       "<g id=\"node19\" class=\"node\">\n",
       "<title>1745401741168</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"2188.89,-66.5 2188.89,-112.5 2331.89,-112.5 2331.89,-66.5 2188.89,-66.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"2260.39\" y=\"-97.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[37.444645 49. ]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"2188.89,-89.5 2331.89,-89.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"2260.39\" y=\"-74.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[0.5 0.5]]</text>\n",
       "</g>\n",
       "<!-- 1745401741168&#45;&gt;1745401954416sum -->\n",
       "<g id=\"edge16\" class=\"edge\">\n",
       "<title>1745401741168&#45;&gt;1745401954416sum</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2331.9,-89.5C2340.84,-89.5 2349.68,-89.5 2357.75,-89.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"2357.88,-93 2367.88,-89.5 2357.88,-86 2357.88,-93\"/>\n",
       "</g>\n",
       "<!-- 1745401741168**2&#45;&gt;1745401741168 -->\n",
       "<g id=\"edge8\" class=\"edge\">\n",
       "<title>1745401741168**2&#45;&gt;1745401741168</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2153.06,-89.5C2160.71,-89.5 2169.49,-89.5 2178.66,-89.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"2178.73,-93 2188.73,-89.5 2178.73,-86 2178.73,-93\"/>\n",
       "</g>\n",
       "<!-- 1744796844208 -->\n",
       "<g id=\"node21\" class=\"node\">\n",
       "<title>1744796844208</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"1470.89\" cy=\"-27.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"1470.89\" y=\"-23.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">&#45;1</text>\n",
       "</g>\n",
       "<!-- 1744796844208&#45;&gt;1745401740688* -->\n",
       "<g id=\"edge12\" class=\"edge\">\n",
       "<title>1744796844208&#45;&gt;1745401740688*</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1496.53,-33.14C1516.71,-37.76 1545.51,-44.36 1568.05,-49.52\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1567.38,-52.96 1577.91,-51.78 1568.95,-46.14 1567.38,-52.96\"/>\n",
       "</g>\n",
       "<!-- 1745401740688 -->\n",
       "<g id=\"node22\" class=\"node\">\n",
       "<title>1745401740688</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"1666.89,-34.5 1666.89,-80.5 1805.89,-80.5 1805.89,-34.5 1666.89,-34.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"1736.39\" y=\"-65.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[&#45;0.880797 &#45;1. ]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"1666.89,-57.5 1805.89,-57.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"1736.39\" y=\"-42.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[6.119203 7. ]]</text>\n",
       "</g>\n",
       "<!-- 1745401740688&#45;&gt;1745401740928+ -->\n",
       "<g id=\"edge23\" class=\"edge\">\n",
       "<title>1745401740688&#45;&gt;1745401740928+</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1806.08,-74.34C1815.59,-76.68 1825.01,-78.99 1833.52,-81.07\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1832.85,-84.51 1843.39,-83.49 1834.51,-77.71 1832.85,-84.51\"/>\n",
       "</g>\n",
       "<!-- 1745401740688*&#45;&gt;1745401740688 -->\n",
       "<g id=\"edge9\" class=\"edge\">\n",
       "<title>1745401740688*&#45;&gt;1745401740688</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1630.98,-57.5C1638.63,-57.5 1647.42,-57.5 1656.57,-57.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1656.62,-61 1666.62,-57.5 1656.62,-54 1656.62,-61\"/>\n",
       "</g>\n",
       "<!-- 1744875845568 -->\n",
       "<g id=\"node24\" class=\"node\">\n",
       "<title>1744875845568</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"326,-0.5 326,-46.5 427,-46.5 427,-0.5 326,-0.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"376.5\" y=\"-31.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[1. 2. 3.]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"326,-23.5 427,-23.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"376.5\" y=\"-8.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[0. 0. 0.]]</text>\n",
       "</g>\n",
       "<!-- 1744875845568&#45;&gt;1745401739824@ -->\n",
       "<g id=\"edge18\" class=\"edge\">\n",
       "<title>1744875845568&#45;&gt;1745401739824@</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M427.03,-34.24C447.79,-38.75 471.43,-43.88 490.29,-47.97\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"489.71,-51.42 500.22,-50.12 491.19,-44.58 489.71,-51.42\"/>\n",
       "</g>\n",
       "<!-- 1745401740256 -->\n",
       "<g id=\"node25\" class=\"node\">\n",
       "<title>1745401740256</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"1097.99,-64.5 1097.99,-110.5 1246.99,-110.5 1246.99,-64.5 1097.99,-64.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"1172.49\" y=\"-95.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[ 2. 35.]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"1097.99,-87.5 1246.99,-87.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"1172.49\" y=\"-72.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.64247733 0. ]]</text>\n",
       "</g>\n",
       "<!-- 1745401740256&#45;&gt;1745401740448Sigmoid -->\n",
       "<g id=\"edge22\" class=\"edge\">\n",
       "<title>1745401740256&#45;&gt;1745401740448Sigmoid</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1247.34,-87.5C1255.92,-87.5 1264.52,-87.5 1272.68,-87.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1272.75,-91 1282.75,-87.5 1272.75,-84 1272.75,-91\"/>\n",
       "</g>\n",
       "<!-- 1745401740256+&#45;&gt;1745401740256 -->\n",
       "<g id=\"edge10\" class=\"edge\">\n",
       "<title>1745401740256+&#45;&gt;1745401740256</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1062.12,-87.5C1069.74,-87.5 1078.52,-87.5 1087.71,-87.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1087.81,-91 1097.81,-87.5 1087.81,-84 1087.81,-91\"/>\n",
       "</g>\n",
       "</g>\n",
       "</svg>\n"
      ],
      "text/plain": [
       "<graphviz.graphs.Digraph at 0x196620cfd30>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MatrixGrad and PyTorch autograds match for sigmoid(relu(x @ w) + b)\n"
     ]
    }
   ],
   "source": [
    "from matrixgrad import Matrix, sigmoid, relu\n",
    "import numpy as np\n",
    "\n",
    "from graphviz import Digraph\n",
    "def graph(self, details=True):\n",
    "    def trace(root):\n",
    "        nodes, edges = set(), set()\n",
    "        def build(v):\n",
    "            if v not in nodes:\n",
    "                nodes.add(v)\n",
    "                if isinstance(v, Matrix):\n",
    "                    for child in v._prev:\n",
    "                        edges.add((child, v))\n",
    "                        build (child)\n",
    "        build(root)\n",
    "        return nodes, edges\n",
    "    \n",
    "    dot = Digraph(format=\"svg\", graph_attr={\"rankdir\": \"LR\"})\n",
    "    nodes, edges = trace(self)\n",
    "    for n in nodes:\n",
    "        uid = str(id(n))\n",
    "        if not isinstance(n, Matrix):\n",
    "            dot.node(name=uid, label = str(n), shape=\"oval\")\n",
    "        else:\n",
    "            if details:\n",
    "                dot.node(name=uid, label = f\"data {n.data} | grad {n.grad}\", shape=\"record\")\n",
    "            else:\n",
    "                dot.node(name=uid, label = f\"data {n.data.shape}\", shape=\"record\")\n",
    "            if n._op:\n",
    "                dot.node(name = uid + n._op, label = n._op)\n",
    "                dot.edge(uid + n._op, uid)\n",
    "\n",
    "    for n1, n2 in edges:\n",
    "        dot.edge(str(id(n1)), str(id(n2)) + n2._op)\n",
    "\n",
    "    return dot\n",
    "\n",
    "\n",
    "# Do the auto-grad thing with the Matrix class\n",
    "w = Matrix([[1,2,-3],[4,5,6]]).transpose()\n",
    "b = Matrix([2,3])\n",
    "x = Matrix([1,2,3])\n",
    "y = Matrix([7, 8])\n",
    "y_hat = sigmoid(relu(x @ w) + b)\n",
    "loss = ((y-y_hat)**2).sum()/y.shape[1]\n",
    "loss.backward()\n",
    "display (graph(loss))\n",
    "\n",
    "# Now try the same thing with PyTorch\n",
    "import torch\n",
    "t_x = torch.tensor(x.data, dtype=torch.float32)\n",
    "t_w = torch.tensor(w.data, dtype=torch.float32, requires_grad=True)\n",
    "t_b = torch.tensor(b.data, dtype=torch.float32, requires_grad=True)\n",
    "t_y_hat = torch.sigmoid(torch.relu(t_x @ t_w) + t_b)\n",
    "t_loss = torch.nn.MSELoss()(torch.as_tensor(y.data), t_y_hat)\n",
    "t_loss.backward()\n",
    "\n",
    "# Make sure they are the same\n",
    "assert np.allclose(y_hat.data, t_y_hat.data.numpy()), f\"Matrix y_hat {y_hat.data} != torch y_hat {t_y_hat.data}\"\n",
    "assert np.allclose(w.grad, t_w.grad.data.numpy()), f\"Matrix grad {w.grad} != torch grad {t_w.grad}\"\n",
    "assert np.allclose(b.grad, t_b.grad.data.numpy()), f\"Matrix grad {b.grad} != torch grad {t_b.grad}\"\n",
    "\n",
    "print (\"MatrixGrad and PyTorch autograds match for sigmoid(relu(x @ w) + b)\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, let's look at how Modules (build on top of Values) can be built run.\n",
    "We'll visualize the models and gradiants with digraph.\n",
    "We'll compare the forward and backward pass values with those generated by PyTorch to validate the Matrix code."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MatrixGrad and PyTorch forward and backward pass match for mse loss of Matix operations\n",
      "Model: Sequential([Layer(inputs=2, outputs=4), ReLU, Layer(inputs=4, outputs=3)])\n",
      "MatrixGrad and PyTorch forward and backward pass match for mse loss on a sequential model\n"
     ]
    },
    {
     "data": {
      "image/svg+xml": [
       "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n",
       "<!DOCTYPE svg PUBLIC \"-//W3C//DTD SVG 1.1//EN\"\n",
       " \"http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd\">\n",
       "<!-- Generated by graphviz version 2.50.0 (0)\n",
       " -->\n",
       "<!-- Pages: 1 -->\n",
       "<svg width=\"5181pt\" height=\"240pt\"\n",
       " viewBox=\"0.00 0.00 5181.48 239.50\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n",
       "<g id=\"graph0\" class=\"graph\" transform=\"scale(1 1) rotate(0) translate(4 235.5)\">\n",
       "<polygon fill=\"white\" stroke=\"transparent\" points=\"-4,4 -4,-235.5 5177.48,-235.5 5177.48,4 -4,4\"/>\n",
       "<!-- 1744875730480 -->\n",
       "<g id=\"node1\" class=\"node\">\n",
       "<title>1744875730480</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"4840.74\" cy=\"-213.5\" rx=\"87.99\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"4840.74\" y=\"-209.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">0.3333333333333333</text>\n",
       "</g>\n",
       "<!-- 1745565809728* -->\n",
       "<g id=\"node6\" class=\"node\">\n",
       "<title>1745565809728*</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"4991.48\" cy=\"-183.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"4991.48\" y=\"-179.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">*</text>\n",
       "</g>\n",
       "<!-- 1744875730480&#45;&gt;1745565809728* -->\n",
       "<g id=\"edge12\" class=\"edge\">\n",
       "<title>1744875730480&#45;&gt;1745565809728*</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M4903.95,-200.95C4921.55,-197.4 4940.07,-193.67 4955.45,-190.57\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"4956.43,-193.94 4965.54,-188.53 4955.05,-187.08 4956.43,-193.94\"/>\n",
       "</g>\n",
       "<!-- 1745565799456 -->\n",
       "<g id=\"node2\" class=\"node\">\n",
       "<title>1745565799456</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"3123.99,-128.5 3123.99,-174.5 3397.99,-174.5 3397.99,-128.5 3123.99,-128.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"3260.99\" y=\"-159.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[&#45;0.01895345 &#45;0.05765361 0.07208243]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"3123.99,-151.5 3397.99,-151.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"3260.99\" y=\"-136.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.01263563 &#45;0.70510244 &#45;3.0852783 ]]</text>\n",
       "</g>\n",
       "<!-- 1745565810640* -->\n",
       "<g id=\"node25\" class=\"node\">\n",
       "<title>1745565810640*</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"3460.99\" cy=\"-121.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"3460.99\" y=\"-117.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">*</text>\n",
       "</g>\n",
       "<!-- 1745565799456&#45;&gt;1745565810640* -->\n",
       "<g id=\"edge13\" class=\"edge\">\n",
       "<title>1745565799456&#45;&gt;1745565810640*</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M3398.16,-130.87C3407.6,-129.44 3416.52,-128.09 3424.49,-126.88\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"3425.01,-130.34 3434.38,-125.38 3423.96,-123.42 3425.01,-130.34\"/>\n",
       "</g>\n",
       "<!-- 1745565799456+ -->\n",
       "<g id=\"node3\" class=\"node\">\n",
       "<title>1745565799456+</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"3060.99\" cy=\"-151.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"3060.99\" y=\"-147.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">+</text>\n",
       "</g>\n",
       "<!-- 1745565799456+&#45;&gt;1745565799456 -->\n",
       "<g id=\"edge1\" class=\"edge\">\n",
       "<title>1745565799456+&#45;&gt;1745565799456</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M3088.29,-151.5C3095.7,-151.5 3104.33,-151.5 3113.7,-151.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"3113.71,-155 3123.71,-151.5 3113.71,-148 3113.71,-155\"/>\n",
       "</g>\n",
       "<!-- 1745565797872 -->\n",
       "<g id=\"node4\" class=\"node\">\n",
       "<title>1745565797872</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"1680.99,-129.5 1680.99,-175.5 2597.99,-175.5 2597.99,-129.5 1680.99,-129.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"2139.49\" y=\"-160.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[&#45;0.02490345 0.04314603 0.02578692] [ 0.03537614 &#45;0.04340858 0.03606645] [ 0.04489662 0.01913276 0.00944839] [ 0.0252091 0.03567517 0.01515545]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"1680.99,-152.5 2597.99,-152.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"2139.49\" y=\"-137.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.0012374 &#45;0.06905046 &#45;0.30214033] [ 0. 0. 0. ] [&#45;0.00175389 &#45;0.09787186 &#45;0.42825255] [&#45;0.00091963 &#45;0.05131766 &#45;0.22454788]]</text>\n",
       "</g>\n",
       "<!-- 1745565799216@ -->\n",
       "<g id=\"node21\" class=\"node\">\n",
       "<title>1745565799216@</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"2660.99\" cy=\"-119.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"2660.99\" y=\"-115.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">@</text>\n",
       "</g>\n",
       "<!-- 1745565797872&#45;&gt;1745565799216@ -->\n",
       "<g id=\"edge16\" class=\"edge\">\n",
       "<title>1745565797872&#45;&gt;1745565799216@</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2502.72,-129.49C2551.78,-126.37 2595.03,-123.63 2623.7,-121.81\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"2624.19,-125.28 2633.95,-121.15 2623.75,-118.3 2624.19,-125.28\"/>\n",
       "</g>\n",
       "<!-- 1745565809728 -->\n",
       "<g id=\"node5\" class=\"node\">\n",
       "<title>1745565809728</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"5054.48,-160.5 5054.48,-206.5 5173.48,-206.5 5173.48,-160.5 5054.48,-160.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"5113.98\" y=\"-191.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[7.5122027]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"5054.48,-183.5 5173.48,-183.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"5113.98\" y=\"-168.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad 1</text>\n",
       "</g>\n",
       "<!-- 1745565809728*&#45;&gt;1745565809728 -->\n",
       "<g id=\"edge2\" class=\"edge\">\n",
       "<title>1745565809728*&#45;&gt;1745565809728</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M5018.63,-183.5C5026.27,-183.5 5035.01,-183.5 5044.01,-183.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"5044.24,-187 5054.24,-183.5 5044.24,-180 5044.24,-187\"/>\n",
       "</g>\n",
       "<!-- 1745565797440 -->\n",
       "<g id=\"node7\" class=\"node\">\n",
       "<title>1745565797440</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"741,-97.5 741,-143.5 1080,-143.5 1080,-97.5 741,-97.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"910.5\" y=\"-128.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[&#45;0.04124644 &#45;0.08839471 0.04382943 0.08906543]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"741,-120.5 1080,-120.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"910.5\" y=\"-105.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.10966751 0. &#45;0.04320876 &#45;0.07223196]]</text>\n",
       "</g>\n",
       "<!-- 1745565798784+ -->\n",
       "<g id=\"node23\" class=\"node\">\n",
       "<title>1745565798784+</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"1143\" cy=\"-87.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"1143\" y=\"-83.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">+</text>\n",
       "</g>\n",
       "<!-- 1745565797440&#45;&gt;1745565798784+ -->\n",
       "<g id=\"edge11\" class=\"edge\">\n",
       "<title>1745565797440&#45;&gt;1745565798784+</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1072.22,-97.49C1084.78,-95.69 1096.48,-94.02 1106.58,-92.57\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1107.09,-96.03 1116.49,-91.15 1106.09,-89.11 1107.09,-96.03\"/>\n",
       "</g>\n",
       "<!-- 1745565798016 -->\n",
       "<g id=\"node8\" class=\"node\">\n",
       "<title>1745565798016</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"2723.99,-161.5 2723.99,-207.5 2997.99,-207.5 2997.99,-161.5 2723.99,-161.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"2860.99\" y=\"-192.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[&#45;0.02458127 &#45;0.06713106 0.06714262]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"2723.99,-184.5 2997.99,-184.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"2860.99\" y=\"-169.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.01263563 &#45;0.70510244 &#45;3.0852783 ]]</text>\n",
       "</g>\n",
       "<!-- 1745565798016&#45;&gt;1745565799456+ -->\n",
       "<g id=\"edge27\" class=\"edge\">\n",
       "<title>1745565798016&#45;&gt;1745565799456+</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2998.16,-161.81C3007.7,-160.22 3016.7,-158.72 3024.72,-157.38\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"3025.39,-160.82 3034.68,-155.72 3024.24,-153.91 3025.39,-160.82\"/>\n",
       "</g>\n",
       "<!-- 1745565798544 -->\n",
       "<g id=\"node9\" class=\"node\">\n",
       "<title>1745565798544</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"741.5,-32.5 741.5,-78.5 1079.5,-78.5 1079.5,-32.5 741.5,-32.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"910.5\" y=\"-63.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[ 0.13917613 0.00028125 0.09497574 &#45;0.01628499]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"741.5,-55.5 1079.5,-55.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"910.5\" y=\"-40.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.10966751 0. &#45;0.04320876 &#45;0.07223196]]</text>\n",
       "</g>\n",
       "<!-- 1745565798544&#45;&gt;1745565798784+ -->\n",
       "<g id=\"edge26\" class=\"edge\">\n",
       "<title>1745565798544&#45;&gt;1745565798784+</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1077.18,-78.5C1087.76,-79.97 1097.64,-81.34 1106.34,-82.55\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1105.94,-86.03 1116.33,-83.94 1106.9,-79.09 1105.94,-86.03\"/>\n",
       "</g>\n",
       "<!-- 1745565798544@ -->\n",
       "<g id=\"node10\" class=\"node\">\n",
       "<title>1745565798544@</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"678\" cy=\"-55.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"678\" y=\"-51.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">@</text>\n",
       "</g>\n",
       "<!-- 1745565798544@&#45;&gt;1745565798544 -->\n",
       "<g id=\"edge3\" class=\"edge\">\n",
       "<title>1745565798544@&#45;&gt;1745565798544</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M705.44,-55.5C712.9,-55.5 721.62,-55.5 731.18,-55.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"731.42,-59 741.42,-55.5 731.42,-52 731.42,-59\"/>\n",
       "</g>\n",
       "<!-- 1745565799072 -->\n",
       "<g id=\"node11\" class=\"node\">\n",
       "<title>1745565799072</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"1964.49,-64.5 1964.49,-110.5 2314.49,-110.5 2314.49,-64.5 1964.49,-64.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"2139.49\" y=\"-95.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[0.09792969 0. 0.13880517 0.07278043]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"1964.49,-87.5 2314.49,-87.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"2139.49\" y=\"-72.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.10966751 &#45;0.08111455 &#45;0.04320876 &#45;0.07223196]]</text>\n",
       "</g>\n",
       "<!-- 1745565799072&#45;&gt;1745565799216@ -->\n",
       "<g id=\"edge22\" class=\"edge\">\n",
       "<title>1745565799072&#45;&gt;1745565799216@</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2314.56,-98.22C2425.5,-105.06 2559.92,-113.34 2623.41,-117.25\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"2623.5,-120.76 2633.7,-117.88 2623.93,-113.77 2623.5,-120.76\"/>\n",
       "</g>\n",
       "<!-- 1745565799072ReLU -->\n",
       "<g id=\"node12\" class=\"node\">\n",
       "<title>1745565799072ReLU</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"1612.5\" cy=\"-87.5\" rx=\"32.49\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"1612.5\" y=\"-83.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">ReLU</text>\n",
       "</g>\n",
       "<!-- 1745565799072ReLU&#45;&gt;1745565799072 -->\n",
       "<g id=\"edge4\" class=\"edge\">\n",
       "<title>1745565799072ReLU&#45;&gt;1745565799072</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1645.16,-87.5C1705.36,-87.5 1840.48,-87.5 1954.11,-87.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1954.34,-91 1964.34,-87.5 1954.34,-84 1954.34,-91\"/>\n",
       "</g>\n",
       "<!-- 1745565827232 -->\n",
       "<g id=\"node13\" class=\"node\">\n",
       "<title>1745565827232</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"205.5,-65.5 205.5,-111.5 409.5,-111.5 409.5,-65.5 205.5,-65.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"307.5\" y=\"-96.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[1. 2.]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"205.5,-88.5 409.5,-88.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"307.5\" y=\"-73.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.00480842 &#45;0.00669109]]</text>\n",
       "</g>\n",
       "<!-- 1745565827232&#45;&gt;1745565798544@ -->\n",
       "<g id=\"edge21\" class=\"edge\">\n",
       "<title>1745565827232&#45;&gt;1745565798544@</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M409.68,-79.44C486.43,-72.57 587.49,-63.52 640.67,-58.75\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"641.31,-62.21 650.96,-57.83 640.69,-55.24 641.31,-62.21\"/>\n",
       "</g>\n",
       "<!-- 1745565810400 -->\n",
       "<g id=\"node14\" class=\"node\">\n",
       "<title>1745565810400</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"3919.99,-130.5 3919.99,-176.5 4179.99,-176.5 4179.99,-130.5 3919.99,-130.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"4049.99\" y=\"-161.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[0.01895345 1.0576537 4.6279173 ]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"3919.99,-153.5 4179.99,-153.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"4049.99\" y=\"-138.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[0.01263563 0.70510244 3.0852783 ]]</text>\n",
       "</g>\n",
       "<!-- 1745565810160**2 -->\n",
       "<g id=\"node28\" class=\"node\">\n",
       "<title>1745565810160**2</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"4242.99\" cy=\"-153.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"4242.99\" y=\"-149.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">**2</text>\n",
       "</g>\n",
       "<!-- 1745565810400&#45;&gt;1745565810160**2 -->\n",
       "<g id=\"edge24\" class=\"edge\">\n",
       "<title>1745565810400&#45;&gt;1745565810160**2</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M4180.25,-153.5C4189.33,-153.5 4197.96,-153.5 4205.71,-153.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"4205.72,-157 4215.72,-153.5 4205.72,-150 4205.72,-157\"/>\n",
       "</g>\n",
       "<!-- 1745565810400+ -->\n",
       "<g id=\"node15\" class=\"node\">\n",
       "<title>1745565810400+</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"3856.99\" cy=\"-153.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"3856.99\" y=\"-149.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">+</text>\n",
       "</g>\n",
       "<!-- 1745565810400+&#45;&gt;1745565810400 -->\n",
       "<g id=\"edge5\" class=\"edge\">\n",
       "<title>1745565810400+&#45;&gt;1745565810400</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M3884.15,-153.5C3891.58,-153.5 3900.23,-153.5 3909.6,-153.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"3909.6,-157 3919.6,-153.5 3909.6,-150 3909.6,-157\"/>\n",
       "</g>\n",
       "<!-- 1745565797632 -->\n",
       "<g id=\"node16\" class=\"node\">\n",
       "<title>1745565797632</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"3528.99,-163.5 3528.99,-209.5 3788.99,-209.5 3788.99,-163.5 3528.99,-163.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"3658.99\" y=\"-194.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[0. 1. 4.7]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"3528.99,-186.5 3788.99,-186.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"3658.99\" y=\"-171.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[0.01263563 0.70510244 3.0852783 ]]</text>\n",
       "</g>\n",
       "<!-- 1745565797632&#45;&gt;1745565810400+ -->\n",
       "<g id=\"edge19\" class=\"edge\">\n",
       "<title>1745565797632&#45;&gt;1745565810400+</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M3789.09,-164.76C3800.44,-162.85 3811.17,-161.05 3820.58,-159.46\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"3821.32,-162.89 3830.6,-157.77 3820.16,-155.98 3821.32,-162.89\"/>\n",
       "</g>\n",
       "<!-- 1745565827328 -->\n",
       "<g id=\"node17\" class=\"node\">\n",
       "<title>1745565827328</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"0,-0.5 0,-46.5 615,-46.5 615,-0.5 0,-0.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"307.5\" y=\"-31.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[ 0.01237025 &#45;0.02298057 0.019196 0.03630478] [ 0.06340294 0.01163091 0.03788987 &#45;0.02629489]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"0,-23.5 615,-23.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"307.5\" y=\"-8.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.10966751 0. &#45;0.04320876 &#45;0.07223196] [&#45;0.21933502 0. &#45;0.08641752 &#45;0.14446393]]</text>\n",
       "</g>\n",
       "<!-- 1745565827328&#45;&gt;1745565798544@ -->\n",
       "<g id=\"edge15\" class=\"edge\">\n",
       "<title>1745565827328&#45;&gt;1745565798544@</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M573.37,-46.5C599.46,-48.77 622.86,-50.8 640.8,-52.36\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"640.7,-55.86 650.97,-53.24 641.31,-48.89 640.7,-55.86\"/>\n",
       "</g>\n",
       "<!-- 1745565809968 -->\n",
       "<g id=\"node18\" class=\"node\">\n",
       "<title>1745565809968</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"4777.24,-130.5 4777.24,-176.5 4904.24,-176.5 4904.24,-130.5 4777.24,-130.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"4840.74\" y=\"-161.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[22.536608]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"4777.24,-153.5 4904.24,-153.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"4840.74\" y=\"-138.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[0.33333334]]</text>\n",
       "</g>\n",
       "<!-- 1745565809968&#45;&gt;1745565809728* -->\n",
       "<g id=\"edge25\" class=\"edge\">\n",
       "<title>1745565809968&#45;&gt;1745565809728*</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M4904.39,-166.14C4921.8,-169.65 4940.1,-173.34 4955.33,-176.41\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"4954.83,-179.88 4965.33,-178.43 4956.22,-173.02 4954.83,-179.88\"/>\n",
       "</g>\n",
       "<!-- 1745565809968sum -->\n",
       "<g id=\"node19\" class=\"node\">\n",
       "<title>1745565809968sum</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"4689.99\" cy=\"-153.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"4689.99\" y=\"-149.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">sum</text>\n",
       "</g>\n",
       "<!-- 1745565809968sum&#45;&gt;1745565809968 -->\n",
       "<g id=\"edge6\" class=\"edge\">\n",
       "<title>1745565809968sum&#45;&gt;1745565809968</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M4717.24,-153.5C4731.12,-153.5 4748.91,-153.5 4766.6,-153.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"4766.98,-157 4776.98,-153.5 4766.98,-150 4766.98,-157\"/>\n",
       "</g>\n",
       "<!-- 1745565799216 -->\n",
       "<g id=\"node20\" class=\"node\">\n",
       "<title>1745565799216</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"2723.99,-96.5 2723.99,-142.5 2997.99,-142.5 2997.99,-96.5 2723.99,-96.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"2860.99\" y=\"-127.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[0.00562783 0.00947746 0.00493981]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"2723.99,-119.5 2997.99,-119.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"2860.99\" y=\"-104.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.01263563 &#45;0.70510244 &#45;3.0852783 ]]</text>\n",
       "</g>\n",
       "<!-- 1745565799216&#45;&gt;1745565799456+ -->\n",
       "<g id=\"edge14\" class=\"edge\">\n",
       "<title>1745565799216&#45;&gt;1745565799456+</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2998.16,-141.51C3007.7,-143.05 3016.7,-144.5 3024.72,-145.8\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"3024.25,-149.27 3034.68,-147.41 3025.36,-142.36 3024.25,-149.27\"/>\n",
       "</g>\n",
       "<!-- 1745565799216@&#45;&gt;1745565799216 -->\n",
       "<g id=\"edge7\" class=\"edge\">\n",
       "<title>1745565799216@&#45;&gt;1745565799216</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M2688.29,-119.5C2695.7,-119.5 2704.33,-119.5 2713.7,-119.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"2713.71,-123 2723.71,-119.5 2713.71,-116 2713.71,-123\"/>\n",
       "</g>\n",
       "<!-- 1745565798784 -->\n",
       "<g id=\"node22\" class=\"node\">\n",
       "<title>1745565798784</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"1206,-64.5 1206,-110.5 1544,-110.5 1544,-64.5 1206,-64.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"1375\" y=\"-95.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[ 0.09792969 &#45;0.08811346 0.13880517 0.07278043]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"1206,-87.5 1544,-87.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"1375\" y=\"-72.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[&#45;0.10966751 0. &#45;0.04320876 &#45;0.07223196]]</text>\n",
       "</g>\n",
       "<!-- 1745565798784&#45;&gt;1745565799072ReLU -->\n",
       "<g id=\"edge18\" class=\"edge\">\n",
       "<title>1745565798784&#45;&gt;1745565799072ReLU</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1544.01,-87.5C1553.11,-87.5 1561.71,-87.5 1569.54,-87.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1569.7,-91 1579.7,-87.5 1569.7,-84 1569.7,-91\"/>\n",
       "</g>\n",
       "<!-- 1745565798784+&#45;&gt;1745565798784 -->\n",
       "<g id=\"edge8\" class=\"edge\">\n",
       "<title>1745565798784+&#45;&gt;1745565798784</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M1170.38,-87.5C1177.74,-87.5 1186.32,-87.5 1195.73,-87.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"1195.81,-91 1205.81,-87.5 1195.81,-84 1195.81,-91\"/>\n",
       "</g>\n",
       "<!-- 1745565810640 -->\n",
       "<g id=\"node24\" class=\"node\">\n",
       "<title>1745565810640</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"3523.99,-98.5 3523.99,-144.5 3793.99,-144.5 3793.99,-98.5 3523.99,-98.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"3658.99\" y=\"-129.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[ 0.01895345 0.05765361 &#45;0.07208243]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"3523.99,-121.5 3793.99,-121.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"3658.99\" y=\"-106.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[0.01263563 0.70510244 3.0852783 ]]</text>\n",
       "</g>\n",
       "<!-- 1745565810640&#45;&gt;1745565810400+ -->\n",
       "<g id=\"edge20\" class=\"edge\">\n",
       "<title>1745565810640&#45;&gt;1745565810400+</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M3794.25,-143.42C3803.76,-144.97 3812.74,-146.44 3820.75,-147.75\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"3820.26,-151.21 3830.69,-149.37 3821.39,-144.3 3820.26,-151.21\"/>\n",
       "</g>\n",
       "<!-- 1745565810640*&#45;&gt;1745565810640 -->\n",
       "<g id=\"edge9\" class=\"edge\">\n",
       "<title>1745565810640*&#45;&gt;1745565810640</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M3488.02,-121.5C3495.52,-121.5 3504.27,-121.5 3513.77,-121.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"3513.92,-125 3523.92,-121.5 3513.92,-118 3513.92,-125\"/>\n",
       "</g>\n",
       "<!-- 1744796844208 -->\n",
       "<g id=\"node26\" class=\"node\">\n",
       "<title>1744796844208</title>\n",
       "<ellipse fill=\"none\" stroke=\"black\" cx=\"3260.99\" cy=\"-91.5\" rx=\"27\" ry=\"18\"/>\n",
       "<text text-anchor=\"middle\" x=\"3260.99\" y=\"-87.8\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">&#45;1</text>\n",
       "</g>\n",
       "<!-- 1744796844208&#45;&gt;1745565810640* -->\n",
       "<g id=\"edge23\" class=\"edge\">\n",
       "<title>1744796844208&#45;&gt;1745565810640*</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M3287.47,-95.36C3322.37,-100.65 3385.2,-110.17 3424.64,-116.14\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"3424.2,-119.62 3434.61,-117.65 3425.25,-112.7 3424.2,-119.62\"/>\n",
       "</g>\n",
       "<!-- 1745565810160 -->\n",
       "<g id=\"node27\" class=\"node\">\n",
       "<title>1745565810160</title>\n",
       "<polygon fill=\"none\" stroke=\"black\" points=\"4305.99,-130.5 4305.99,-176.5 4626.99,-176.5 4626.99,-130.5 4305.99,-130.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"4466.49\" y=\"-161.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">data [[3.5923321e&#45;04 1.1186312e+00 2.1417618e+01]]</text>\n",
       "<polyline fill=\"none\" stroke=\"black\" points=\"4305.99,-153.5 4626.99,-153.5 \"/>\n",
       "<text text-anchor=\"middle\" x=\"4466.49\" y=\"-138.3\" font-family=\"Times New Roman,serif\" font-size=\"14.00\">grad [[0.33333334 0.33333334 0.33333334]]</text>\n",
       "</g>\n",
       "<!-- 1745565810160&#45;&gt;1745565809968sum -->\n",
       "<g id=\"edge17\" class=\"edge\">\n",
       "<title>1745565810160&#45;&gt;1745565809968sum</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M4627,-153.5C4636.32,-153.5 4645.07,-153.5 4652.89,-153.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"4652.95,-157 4662.95,-153.5 4652.95,-150 4652.95,-157\"/>\n",
       "</g>\n",
       "<!-- 1745565810160**2&#45;&gt;1745565810160 -->\n",
       "<g id=\"edge10\" class=\"edge\">\n",
       "<title>1745565810160**2&#45;&gt;1745565810160</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M4270.27,-153.5C4277.67,-153.5 4286.31,-153.5 4295.76,-153.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"4295.88,-157 4305.88,-153.5 4295.88,-150 4295.88,-157\"/>\n",
       "</g>\n",
       "</g>\n",
       "</svg>\n"
      ],
      "text/plain": [
       "<graphviz.graphs.Digraph at 0x1966bd47be0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Simple 3-layer model, internally using Value objects as weights for the linear layers\n",
    "from matrixgrad import Linear, ReLU, Sequential, mse_loss\n",
    "import numpy as np\n",
    "\n",
    "\"\"\" Test the classes defined in this file \"\"\"\n",
    "l = Linear(2, 3)\n",
    "x = np.array([[1.0, 2.0]])\n",
    "y = np.array([[0.0, 1.0, 4.7]])\n",
    "yhat = l(x)\n",
    "yhat_relu = relu(yhat)\n",
    "loss = mse_loss([y], [yhat_relu])\n",
    "loss.backward()\n",
    "\n",
    "import torch\n",
    "t_l = torch.nn.Linear(2, 3)\n",
    "t_l.weight.data = torch.tensor(l.w.data.T)\n",
    "t_l.bias.data = torch.tensor(l.b.data)\n",
    "t_x = torch.tensor(x, dtype=torch.float32)\n",
    "t_y = torch.tensor(y, dtype=torch.float32)\n",
    "t_yhat = t_l(t_x)\n",
    "t_yhat_relu = torch.nn.ReLU()(t_yhat)\n",
    "t_loss = torch.nn.MSELoss()(t_y, t_yhat_relu)\n",
    "t_loss.backward()\n",
    "\n",
    "assert np.allclose(yhat.data, t_yhat.data), \"Linear layer output doesn't match PyTorch\"\n",
    "assert np.allclose(yhat_relu.data, t_yhat_relu.data), \"ReLU output doesn't match PyTorch\"\n",
    "assert np.allclose(loss.data, t_loss.data), \"MSE loss doesn't match PyTorch\"\n",
    "assert np.allclose(l.w.grad, t_l.weight.grad.T), \"Linear layer weight gradient doesn't match PyTorch\"\n",
    "assert np.allclose(l.b.grad, t_l.bias.grad), \"Linear layer bias gradient doesn't match PyTorch\"\n",
    "\n",
    "print (\"MatrixGrad and PyTorch forward and backward pass match for mse loss of Matix operations\")\n",
    "\n",
    "# Now lets try the same thing with a Sequential model\n",
    "model = Sequential([Linear(2, 4), ReLU(), Linear(4, 3)])\n",
    "yhat = model(x)\n",
    "loss = mse_loss([y], [yhat])\n",
    "loss.backward()\n",
    "print (f\"Model: {model}\")\n",
    "\n",
    "t_model = torch.nn.Sequential(torch.nn.Linear(2, 4), torch.nn.ReLU(), torch.nn.Linear(4, 2))\n",
    "t_model[0].weight.data = torch.tensor(model.layers[0].w.data.T)\n",
    "t_model[0].bias.data = torch.tensor(model.layers[0].b.data)\n",
    "t_model[2].weight.data = torch.tensor(model.layers[2].w.data.T)\n",
    "t_model[2].bias.data = torch.tensor(model.layers[2].b.data)\n",
    "t_yhat = t_model(t_x)\n",
    "t_loss = torch.nn.MSELoss()(t_y, t_yhat)\n",
    "t_loss.backward()\n",
    "\n",
    "assert np.allclose(yhat.data, t_yhat.data), \"Sequential model output doesn't match PyTorch\"\n",
    "assert np.allclose(loss.data, t_loss.data), \"MSE loss doesn't match PyTorch\"\n",
    "assert np.allclose(model.layers[0].w.grad, t_model[0].weight.grad.T), \"Linear layer weight gradient doesn't match PyTorch\"\n",
    "assert np.allclose(model.layers[0].b.grad, t_model[0].bias.grad), \"Linear layer bias gradient doesn't match PyTorch\"\n",
    "assert np.allclose(model.layers[2].w.grad, t_model[2].weight.grad.T), \"Linear layer weight gradient doesn't match PyTorch\"\n",
    "assert np.allclose(model.layers[2].b.grad, t_model[2].bias.grad), \"Linear layer bias gradient doesn't match PyTorch\"\n",
    "\n",
    "print (\"MatrixGrad and PyTorch forward and backward pass match for mse loss on a sequential model\")\n",
    "display(graph(loss))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, let's try training a simple regression model using the layers that sit on top of Matrix/autograd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training model Sequential([Layer(inputs=2, outputs=3), ReLU, Layer(inputs=3, outputs=1)]) to learn 2*x1+x2+1\n",
      "Epoch 0: avg training loss=1.9284\n",
      "Epoch 1: avg training loss=1.4632\n",
      "Epoch 2: avg training loss=0.3330\n",
      "Epoch 3: avg training loss=1.1135\n",
      "Epoch 4: avg training loss=0.0765\n",
      "Epoch 5: avg training loss=0.0282\n",
      "Epoch 6: avg training loss=0.0190\n",
      "Epoch 7: avg training loss=0.0168\n",
      "Epoch 8: avg training loss=0.0138\n",
      "Epoch 9: avg training loss=0.0117\n",
      "Epoch 10: avg training loss=0.0104\n",
      "Epoch 11: avg training loss=0.0091\n",
      "Epoch 12: avg training loss=0.0083\n",
      "Epoch 13: avg training loss=0.0071\n",
      "Epoch 14: avg training loss=0.0060\n",
      "Epoch 15: avg training loss=0.0054\n",
      "Epoch 16: avg training loss=0.0049\n",
      "Epoch 17: avg training loss=0.0044\n",
      "Epoch 18: avg training loss=0.0040\n",
      "Epoch 19: avg training loss=0.0037\n",
      "Epoch 20: avg training loss=0.0034\n",
      "Epoch 21: avg training loss=0.0031\n",
      "Epoch 22: avg training loss=0.0029\n",
      "Epoch 23: avg training loss=0.0028\n",
      "Epoch 24: avg training loss=0.0026\n",
      "Epoch 25: avg training loss=0.0025\n",
      "Epoch 26: avg training loss=0.0024\n",
      "Epoch 27: avg training loss=0.0023\n",
      "Epoch 28: avg training loss=0.0022\n",
      "Epoch 29: avg training loss=0.0022\n",
      "Epoch 30: avg training loss=0.0021\n",
      "Epoch 31: avg training loss=0.0021\n",
      "Epoch 32: avg training loss=0.0020\n",
      "Epoch 33: avg training loss=0.0020\n",
      "Epoch 34: avg training loss=0.0019\n",
      "Epoch 35: avg training loss=0.0019\n",
      "Epoch 36: avg training loss=0.0019\n",
      "Epoch 37: avg training loss=0.0019\n",
      "Epoch 38: avg training loss=0.0018\n",
      "Epoch 39: avg training loss=0.0018\n",
      "Epoch 40: avg training loss=0.0018\n",
      "Epoch 41: avg training loss=0.0018\n",
      "Epoch 42: avg training loss=0.0018\n",
      "Epoch 43: avg training loss=0.0018\n",
      "Epoch 44: avg training loss=0.0018\n",
      "Epoch 45: avg training loss=0.0017\n",
      "Epoch 46: avg training loss=0.0017\n",
      "Epoch 47: avg training loss=0.0017\n",
      "Epoch 48: avg training loss=0.0017\n",
      "Epoch 49: avg training loss=0.0017\n",
      "SUCCESS: test loss 0.0012154125142842531\n"
     ]
    }
   ],
   "source": [
    "from matrixgrad import Matrix, Sequential, Linear, ReLU, mse_loss, SGD, fit\n",
    "import numpy as np\n",
    "\n",
    "np.random.seed(12)  # 12th man - go Seahawks!\n",
    "\n",
    "\"\"\" Train a model to learn the function f(x1,x2) = x1+x2+1 with SGD and verify the fit \"\"\"\n",
    "\n",
    "# Create the model\n",
    "model = Sequential([Linear(2, 3), ReLU(), Linear(3, 1)])\n",
    "opt = SGD(model.parameters(), lr=.5, lr_decay=.007)\n",
    "loss_fn = mse_loss\n",
    "\n",
    "# Train the model\n",
    "print (f\"Training model {model} to learn 2*x1+x2+1\")\n",
    "x_train = [Matrix([[np.random.uniform(-1, 1), np.random.uniform(-1, 1)]]) for _ in range(128)]\n",
    "y_train = [2*m.data[0,0] + m.data[0,1] + 1 for m in x_train]\n",
    "fit (model, x_train, y_train, loss_fn=loss_fn, optimizer=opt, epochs=50, batch_size=32, verbose=True)\n",
    "\n",
    "# Verify the fit\n",
    "x_test = [Matrix([[np.random.uniform(0, 1), np.random.uniform(0, 1)]]) for _ in range(32)]\n",
    "y_test = [[2*m.data[0,0] + m.data[0,1] + 1] for m in x_test]\n",
    "yhat_test = [model(x_testi) for x_testi in x_test]\n",
    "test_loss = mse_loss(y_test, yhat_test)\n",
    "if test_loss.data[0,0] > .01:\n",
    "    print (f\"FAILED: test loss {test_loss.data[0,0]}\")\n",
    "else:\n",
    "    print (f\"SUCCESS: test loss {test_loss.data[0,0]}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finally, let's train a simple classification problem using the layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0: avg training loss=0.6930, accuracy=0.5\n",
      "Epoch 1: avg training loss=0.6679, accuracy=0.5\n",
      "Epoch 2: avg training loss=0.5998, accuracy=0.79\n",
      "Epoch 3: avg training loss=0.5133, accuracy=0.77\n",
      "Epoch 4: avg training loss=0.4346, accuracy=0.78\n",
      "Epoch 5: avg training loss=0.3810, accuracy=0.8\n",
      "Epoch 6: avg training loss=0.3474, accuracy=0.83\n",
      "Epoch 7: avg training loss=0.3303, accuracy=0.83\n",
      "Epoch 8: avg training loss=0.3198, accuracy=0.86\n",
      "Epoch 9: avg training loss=0.3055, accuracy=0.86\n",
      "Epoch 10: avg training loss=0.2878, accuracy=0.87\n",
      "Epoch 11: avg training loss=0.2770, accuracy=0.86\n",
      "Epoch 12: avg training loss=0.2834, accuracy=0.85\n",
      "Epoch 13: avg training loss=0.3005, accuracy=0.84\n",
      "Epoch 14: avg training loss=0.3105, accuracy=0.84\n",
      "Epoch 15: avg training loss=0.3087, accuracy=0.85\n",
      "Epoch 16: avg training loss=0.3008, accuracy=0.83\n",
      "Epoch 17: avg training loss=0.2939, accuracy=0.85\n",
      "Epoch 18: avg training loss=0.2916, accuracy=0.85\n",
      "Epoch 19: avg training loss=0.2926, accuracy=0.86\n",
      "Epoch 20: avg training loss=0.2930, accuracy=0.87\n",
      "Epoch 21: avg training loss=0.2905, accuracy=0.87\n",
      "Epoch 22: avg training loss=0.2854, accuracy=0.87\n",
      "Epoch 23: avg training loss=0.2794, accuracy=0.87\n",
      "Epoch 24: avg training loss=0.2741, accuracy=0.87\n",
      "Epoch 25: avg training loss=0.2715, accuracy=0.85\n",
      "Epoch 26: avg training loss=0.2722, accuracy=0.84\n",
      "Epoch 27: avg training loss=0.2749, accuracy=0.84\n",
      "Epoch 28: avg training loss=0.2775, accuracy=0.85\n",
      "Epoch 29: avg training loss=0.2789, accuracy=0.85\n",
      "Epoch 30: avg training loss=0.2785, accuracy=0.85\n",
      "Epoch 31: avg training loss=0.2768, accuracy=0.85\n",
      "Epoch 32: avg training loss=0.2754, accuracy=0.85\n",
      "Epoch 33: avg training loss=0.2748, accuracy=0.85\n",
      "Epoch 34: avg training loss=0.2747, accuracy=0.85\n",
      "Epoch 35: avg training loss=0.2750, accuracy=0.86\n",
      "Epoch 36: avg training loss=0.2751, accuracy=0.86\n",
      "Epoch 37: avg training loss=0.2733, accuracy=0.86\n",
      "Epoch 38: avg training loss=0.2691, accuracy=0.86\n",
      "Epoch 39: avg training loss=0.2645, accuracy=0.87\n",
      "Epoch 40: avg training loss=0.2617, accuracy=0.86\n",
      "Epoch 41: avg training loss=0.2570, accuracy=0.86\n",
      "Epoch 42: avg training loss=0.2550, accuracy=0.87\n",
      "Epoch 43: avg training loss=0.2510, accuracy=0.86\n",
      "Epoch 44: avg training loss=0.2493, accuracy=0.86\n",
      "Epoch 45: avg training loss=0.2456, accuracy=0.86\n",
      "Epoch 46: avg training loss=0.2426, accuracy=0.88\n",
      "Epoch 47: avg training loss=0.2386, accuracy=0.88\n",
      "Epoch 48: avg training loss=0.2378, accuracy=0.88\n",
      "Epoch 49: avg training loss=0.2344, accuracy=0.88\n",
      "Epoch 50: avg training loss=0.2337, accuracy=0.88\n",
      "Epoch 51: avg training loss=0.2329, accuracy=0.88\n",
      "Epoch 52: avg training loss=0.2322, accuracy=0.88\n",
      "Epoch 53: avg training loss=0.2313, accuracy=0.88\n",
      "Epoch 54: avg training loss=0.2305, accuracy=0.88\n",
      "Epoch 55: avg training loss=0.2304, accuracy=0.88\n",
      "Epoch 56: avg training loss=0.2284, accuracy=0.89\n",
      "Epoch 57: avg training loss=0.2233, accuracy=0.89\n",
      "Epoch 58: avg training loss=0.2155, accuracy=0.89\n",
      "Epoch 59: avg training loss=0.2078, accuracy=0.88\n",
      "Epoch 60: avg training loss=0.1991, accuracy=0.88\n",
      "Epoch 61: avg training loss=0.1896, accuracy=0.89\n",
      "Epoch 62: avg training loss=0.1741, accuracy=0.9\n",
      "Epoch 63: avg training loss=0.1572, accuracy=0.92\n",
      "Epoch 64: avg training loss=0.1417, accuracy=0.93\n",
      "Epoch 65: avg training loss=0.1260, accuracy=0.95\n",
      "Epoch 66: avg training loss=0.1110, accuracy=0.98\n",
      "Epoch 67: avg training loss=0.0981, accuracy=0.98\n",
      "Epoch 68: avg training loss=0.0842, accuracy=0.99\n",
      "Epoch 69: avg training loss=0.0738, accuracy=0.99\n",
      "Epoch 70: avg training loss=0.0640, accuracy=0.99\n",
      "Epoch 71: avg training loss=0.0569, accuracy=0.99\n",
      "Epoch 72: avg training loss=0.0506, accuracy=0.99\n",
      "Epoch 73: avg training loss=0.0440, accuracy=0.99\n",
      "Epoch 74: avg training loss=0.0377, accuracy=1.0\n",
      "Epoch 75: avg training loss=0.0314, accuracy=1.0\n",
      "Epoch 76: avg training loss=0.0258, accuracy=1.0\n",
      "Epoch 77: avg training loss=0.0215, accuracy=1.0\n",
      "Epoch 78: avg training loss=0.0179, accuracy=1.0\n",
      "Epoch 79: avg training loss=0.0151, accuracy=1.0\n",
      "Epoch 80: avg training loss=0.0133, accuracy=1.0\n",
      "Epoch 81: avg training loss=0.0115, accuracy=1.0\n",
      "Epoch 82: avg training loss=0.0099, accuracy=1.0\n",
      "Epoch 83: avg training loss=0.0088, accuracy=1.0\n",
      "Epoch 84: avg training loss=0.0077, accuracy=1.0\n",
      "Epoch 85: avg training loss=0.0064, accuracy=1.0\n",
      "Epoch 86: avg training loss=0.0053, accuracy=1.0\n",
      "Epoch 87: avg training loss=0.0045, accuracy=1.0\n",
      "Epoch 88: avg training loss=0.0039, accuracy=1.0\n",
      "Epoch 89: avg training loss=0.0033, accuracy=1.0\n",
      "Epoch 90: avg training loss=0.0029, accuracy=1.0\n",
      "Epoch 91: avg training loss=0.0026, accuracy=1.0\n",
      "Epoch 92: avg training loss=0.0024, accuracy=1.0\n",
      "Epoch 93: avg training loss=0.0022, accuracy=1.0\n",
      "Epoch 94: avg training loss=0.0020, accuracy=1.0\n",
      "Epoch 95: avg training loss=0.0019, accuracy=1.0\n",
      "Epoch 96: avg training loss=0.0018, accuracy=1.0\n",
      "Epoch 97: avg training loss=0.0017, accuracy=1.0\n",
      "Epoch 98: avg training loss=0.0017, accuracy=1.0\n",
      "Epoch 99: avg training loss=0.0016, accuracy=1.0\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matrixgrad import *\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.datasets import make_moons\n",
    "\n",
    "np.random.seed(12)  # 12th man, go Seahawks!\n",
    "\n",
    "# Build the model\n",
    "model = Sequential([Linear(2, 10), ReLU(), Linear(10, 1), Sigmoid()])\n",
    "opt = Adam(model.parameters(), lr=.15, verbose=False)\n",
    "loss_fn = binary_cross_entryopy_loss\n",
    "\n",
    "# Train it\n",
    "x, y = make_moons(n_samples=100, noise=0.1) # Two interlevered half circles\n",
    "y = y.reshape ((100, 1, 1))                 # Make y an array of 1x1 arrays to match the output shape of the model\n",
    "fit (model, x, y, loss_fn=loss_fn, optimizer=opt, epochs=100, batch_size=None, metrics=(accuracy,), verbose=True)\n",
    "\n",
    "# Visualize the decision boundary\n",
    "x1 = np.linspace(-2, 2.5, 100)\n",
    "x2 = np.linspace(-2, 2, 100)\n",
    "xx1, xx2 = np.meshgrid(x1, x2)\n",
    "X = np.hstack((xx1.reshape(-1,1), xx2.reshape(-1,1)))\n",
    "yhat = model(X).data\n",
    "plt.contourf(x1, x2, np.round(yhat.reshape(100,100)), cmap='jet', alpha=.5)\n",
    "plt.scatter(x[:,0], x[:,1], c=y, s=20, cmap='jet')\n",
    "plt.title (\"Matrix classification decision boundary\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "PyTorch and TensorFLow versions of the above, for comparison..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0: loss=0.7579, accuracy=29%\n",
      "Epoch 1: loss=0.6900, accuracy=50%\n",
      "Epoch 2: loss=0.6592, accuracy=60%\n",
      "Epoch 3: loss=0.6251, accuracy=79%\n",
      "Epoch 4: loss=0.5783, accuracy=84%\n",
      "Epoch 5: loss=0.5201, accuracy=85%\n",
      "Epoch 6: loss=0.4559, accuracy=82%\n",
      "Epoch 7: loss=0.3968, accuracy=82%\n",
      "Epoch 8: loss=0.3558, accuracy=82%\n",
      "Epoch 9: loss=0.3364, accuracy=82%\n",
      "Epoch 10: loss=0.3309, accuracy=82%\n",
      "Epoch 11: loss=0.3279, accuracy=82%\n",
      "Epoch 12: loss=0.3220, accuracy=82%\n",
      "Epoch 13: loss=0.3128, accuracy=85%\n",
      "Epoch 14: loss=0.3017, accuracy=85%\n",
      "Epoch 15: loss=0.2902, accuracy=85%\n",
      "Epoch 16: loss=0.2792, accuracy=83%\n",
      "Epoch 17: loss=0.2682, accuracy=85%\n",
      "Epoch 18: loss=0.2599, accuracy=88%\n",
      "Epoch 19: loss=0.2569, accuracy=87%\n",
      "Epoch 20: loss=0.2590, accuracy=86%\n",
      "Epoch 21: loss=0.2606, accuracy=88%\n",
      "Epoch 22: loss=0.2576, accuracy=88%\n",
      "Epoch 23: loss=0.2513, accuracy=87%\n",
      "Epoch 24: loss=0.2436, accuracy=88%\n",
      "Epoch 25: loss=0.2349, accuracy=88%\n",
      "Epoch 26: loss=0.2249, accuracy=88%\n",
      "Epoch 27: loss=0.2160, accuracy=89%\n",
      "Epoch 28: loss=0.2097, accuracy=90%\n",
      "Epoch 29: loss=0.2057, accuracy=89%\n",
      "Epoch 30: loss=0.2021, accuracy=89%\n",
      "Epoch 31: loss=0.1983, accuracy=90%\n",
      "Epoch 32: loss=0.1949, accuracy=89%\n",
      "Epoch 33: loss=0.1922, accuracy=90%\n",
      "Epoch 34: loss=0.1896, accuracy=90%\n",
      "Epoch 35: loss=0.1859, accuracy=90%\n",
      "Epoch 36: loss=0.1818, accuracy=91%\n",
      "Epoch 37: loss=0.1783, accuracy=91%\n",
      "Epoch 38: loss=0.1746, accuracy=93%\n",
      "Epoch 39: loss=0.1703, accuracy=93%\n",
      "Epoch 40: loss=0.1661, accuracy=93%\n",
      "Epoch 41: loss=0.1621, accuracy=93%\n",
      "Epoch 42: loss=0.1579, accuracy=93%\n",
      "Epoch 43: loss=0.1531, accuracy=94%\n",
      "Epoch 44: loss=0.1478, accuracy=94%\n",
      "Epoch 45: loss=0.1425, accuracy=95%\n",
      "Epoch 46: loss=0.1370, accuracy=95%\n",
      "Epoch 47: loss=0.1314, accuracy=95%\n",
      "Epoch 48: loss=0.1257, accuracy=95%\n",
      "Epoch 49: loss=0.1202, accuracy=96%\n",
      "Epoch 50: loss=0.1152, accuracy=96%\n",
      "Epoch 51: loss=0.1101, accuracy=97%\n",
      "Epoch 52: loss=0.1049, accuracy=97%\n",
      "Epoch 53: loss=0.0996, accuracy=98%\n",
      "Epoch 54: loss=0.0944, accuracy=98%\n",
      "Epoch 55: loss=0.0891, accuracy=98%\n",
      "Epoch 56: loss=0.0839, accuracy=98%\n",
      "Epoch 57: loss=0.0789, accuracy=98%\n",
      "Epoch 58: loss=0.0743, accuracy=98%\n",
      "Epoch 59: loss=0.0698, accuracy=98%\n",
      "Epoch 60: loss=0.0657, accuracy=98%\n",
      "Epoch 61: loss=0.0619, accuracy=99%\n",
      "Epoch 62: loss=0.0583, accuracy=99%\n",
      "Epoch 63: loss=0.0548, accuracy=99%\n",
      "Epoch 64: loss=0.0516, accuracy=99%\n",
      "Epoch 65: loss=0.0486, accuracy=100%\n",
      "Epoch 66: loss=0.0458, accuracy=100%\n",
      "Epoch 67: loss=0.0432, accuracy=100%\n",
      "Epoch 68: loss=0.0407, accuracy=100%\n",
      "Epoch 69: loss=0.0384, accuracy=100%\n",
      "Epoch 70: loss=0.0363, accuracy=100%\n",
      "Epoch 71: loss=0.0342, accuracy=100%\n",
      "Epoch 72: loss=0.0322, accuracy=100%\n",
      "Epoch 73: loss=0.0303, accuracy=100%\n",
      "Epoch 74: loss=0.0285, accuracy=100%\n",
      "Epoch 75: loss=0.0269, accuracy=100%\n",
      "Epoch 76: loss=0.0254, accuracy=100%\n",
      "Epoch 77: loss=0.0242, accuracy=100%\n",
      "Epoch 78: loss=0.0230, accuracy=100%\n",
      "Epoch 79: loss=0.0219, accuracy=100%\n",
      "Epoch 80: loss=0.0209, accuracy=100%\n",
      "Epoch 81: loss=0.0198, accuracy=100%\n",
      "Epoch 82: loss=0.0189, accuracy=100%\n",
      "Epoch 83: loss=0.0181, accuracy=100%\n",
      "Epoch 84: loss=0.0173, accuracy=100%\n",
      "Epoch 85: loss=0.0166, accuracy=100%\n",
      "Epoch 86: loss=0.0159, accuracy=100%\n",
      "Epoch 87: loss=0.0153, accuracy=100%\n",
      "Epoch 88: loss=0.0147, accuracy=100%\n",
      "Epoch 89: loss=0.0142, accuracy=100%\n",
      "Epoch 90: loss=0.0137, accuracy=100%\n",
      "Epoch 91: loss=0.0132, accuracy=100%\n",
      "Epoch 92: loss=0.0128, accuracy=100%\n",
      "Epoch 93: loss=0.0123, accuracy=100%\n",
      "Epoch 94: loss=0.0119, accuracy=100%\n",
      "Epoch 95: loss=0.0116, accuracy=100%\n",
      "Epoch 96: loss=0.0112, accuracy=100%\n",
      "Epoch 97: loss=0.0109, accuracy=100%\n",
      "Epoch 98: loss=0.0106, accuracy=100%\n",
      "Epoch 99: loss=0.0103, accuracy=100%\n",
      "Epoch 100: loss=0.0100, accuracy=100%\n",
      "Epoch 101: loss=0.0097, accuracy=100%\n",
      "Epoch 102: loss=0.0095, accuracy=100%\n",
      "Epoch 103: loss=0.0092, accuracy=100%\n",
      "Epoch 104: loss=0.0090, accuracy=100%\n",
      "Epoch 105: loss=0.0088, accuracy=100%\n",
      "Epoch 106: loss=0.0086, accuracy=100%\n",
      "Epoch 107: loss=0.0083, accuracy=100%\n",
      "Epoch 108: loss=0.0081, accuracy=100%\n",
      "Epoch 109: loss=0.0079, accuracy=100%\n",
      "Epoch 110: loss=0.0078, accuracy=100%\n",
      "Epoch 111: loss=0.0076, accuracy=100%\n",
      "Epoch 112: loss=0.0074, accuracy=100%\n",
      "Epoch 113: loss=0.0072, accuracy=100%\n",
      "Epoch 114: loss=0.0071, accuracy=100%\n",
      "Epoch 115: loss=0.0069, accuracy=100%\n",
      "Epoch 116: loss=0.0068, accuracy=100%\n",
      "Epoch 117: loss=0.0066, accuracy=100%\n",
      "Epoch 118: loss=0.0065, accuracy=100%\n",
      "Epoch 119: loss=0.0064, accuracy=100%\n",
      "Epoch 120: loss=0.0063, accuracy=100%\n",
      "Epoch 121: loss=0.0061, accuracy=100%\n",
      "Epoch 122: loss=0.0060, accuracy=100%\n",
      "Epoch 123: loss=0.0059, accuracy=100%\n",
      "Epoch 124: loss=0.0058, accuracy=100%\n",
      "Epoch 125: loss=0.0057, accuracy=100%\n",
      "Epoch 126: loss=0.0056, accuracy=100%\n",
      "Epoch 127: loss=0.0055, accuracy=100%\n",
      "Epoch 128: loss=0.0054, accuracy=100%\n",
      "Epoch 129: loss=0.0053, accuracy=100%\n",
      "Epoch 130: loss=0.0052, accuracy=100%\n",
      "Epoch 131: loss=0.0051, accuracy=100%\n",
      "Epoch 132: loss=0.0051, accuracy=100%\n",
      "Epoch 133: loss=0.0050, accuracy=100%\n",
      "Epoch 134: loss=0.0049, accuracy=100%\n",
      "Epoch 135: loss=0.0048, accuracy=100%\n",
      "Epoch 136: loss=0.0048, accuracy=100%\n",
      "Epoch 137: loss=0.0047, accuracy=100%\n",
      "Epoch 138: loss=0.0046, accuracy=100%\n",
      "Epoch 139: loss=0.0045, accuracy=100%\n",
      "Epoch 140: loss=0.0045, accuracy=100%\n",
      "Epoch 141: loss=0.0044, accuracy=100%\n",
      "Epoch 142: loss=0.0043, accuracy=100%\n",
      "Epoch 143: loss=0.0043, accuracy=100%\n",
      "Epoch 144: loss=0.0042, accuracy=100%\n",
      "Epoch 145: loss=0.0042, accuracy=100%\n",
      "Epoch 146: loss=0.0041, accuracy=100%\n",
      "Epoch 147: loss=0.0041, accuracy=100%\n",
      "Epoch 148: loss=0.0040, accuracy=100%\n",
      "Epoch 149: loss=0.0039, accuracy=100%\n",
      "Epoch 150: loss=0.0039, accuracy=100%\n",
      "Epoch 151: loss=0.0038, accuracy=100%\n",
      "Epoch 152: loss=0.0038, accuracy=100%\n",
      "Epoch 153: loss=0.0037, accuracy=100%\n",
      "Epoch 154: loss=0.0037, accuracy=100%\n",
      "Epoch 155: loss=0.0036, accuracy=100%\n",
      "Epoch 156: loss=0.0036, accuracy=100%\n",
      "Epoch 157: loss=0.0035, accuracy=100%\n",
      "Epoch 158: loss=0.0035, accuracy=100%\n",
      "Epoch 159: loss=0.0035, accuracy=100%\n",
      "Epoch 160: loss=0.0034, accuracy=100%\n",
      "Epoch 161: loss=0.0034, accuracy=100%\n",
      "Epoch 162: loss=0.0033, accuracy=100%\n",
      "Epoch 163: loss=0.0033, accuracy=100%\n",
      "Epoch 164: loss=0.0033, accuracy=100%\n",
      "Epoch 165: loss=0.0032, accuracy=100%\n",
      "Epoch 166: loss=0.0032, accuracy=100%\n",
      "Epoch 167: loss=0.0031, accuracy=100%\n",
      "Epoch 168: loss=0.0031, accuracy=100%\n",
      "Epoch 169: loss=0.0031, accuracy=100%\n",
      "Epoch 170: loss=0.0030, accuracy=100%\n",
      "Epoch 171: loss=0.0030, accuracy=100%\n",
      "Epoch 172: loss=0.0030, accuracy=100%\n",
      "Epoch 173: loss=0.0029, accuracy=100%\n",
      "Epoch 174: loss=0.0029, accuracy=100%\n",
      "Epoch 175: loss=0.0029, accuracy=100%\n",
      "Epoch 176: loss=0.0028, accuracy=100%\n",
      "Epoch 177: loss=0.0028, accuracy=100%\n",
      "Epoch 178: loss=0.0028, accuracy=100%\n",
      "Epoch 179: loss=0.0027, accuracy=100%\n",
      "Epoch 180: loss=0.0027, accuracy=100%\n",
      "Epoch 181: loss=0.0027, accuracy=100%\n",
      "Epoch 182: loss=0.0027, accuracy=100%\n",
      "Epoch 183: loss=0.0026, accuracy=100%\n",
      "Epoch 184: loss=0.0026, accuracy=100%\n",
      "Epoch 185: loss=0.0026, accuracy=100%\n",
      "Epoch 186: loss=0.0025, accuracy=100%\n",
      "Epoch 187: loss=0.0025, accuracy=100%\n",
      "Epoch 188: loss=0.0025, accuracy=100%\n",
      "Epoch 189: loss=0.0025, accuracy=100%\n",
      "Epoch 190: loss=0.0024, accuracy=100%\n",
      "Epoch 191: loss=0.0024, accuracy=100%\n",
      "Epoch 192: loss=0.0024, accuracy=100%\n",
      "Epoch 193: loss=0.0024, accuracy=100%\n",
      "Epoch 194: loss=0.0023, accuracy=100%\n",
      "Epoch 195: loss=0.0023, accuracy=100%\n",
      "Epoch 196: loss=0.0023, accuracy=100%\n",
      "Epoch 197: loss=0.0023, accuracy=100%\n",
      "Epoch 198: loss=0.0023, accuracy=100%\n",
      "Epoch 199: loss=0.0022, accuracy=100%\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/200\n",
      "4/4 [==============================] - 0s 7ms/step - loss: 0.6447 - accuracy: 0.6700\n",
      "Epoch 2/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.3535 - accuracy: 0.8300\n",
      "Epoch 3/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2941 - accuracy: 0.8700\n",
      "Epoch 4/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2818 - accuracy: 0.9000\n",
      "Epoch 5/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2518 - accuracy: 0.8800\n",
      "Epoch 6/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2495 - accuracy: 0.8900\n",
      "Epoch 7/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2671 - accuracy: 0.8600\n",
      "Epoch 8/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2519 - accuracy: 0.9000\n",
      "Epoch 9/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2764 - accuracy: 0.8800\n",
      "Epoch 10/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2517 - accuracy: 0.8800\n",
      "Epoch 11/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2494 - accuracy: 0.8700\n",
      "Epoch 12/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2399 - accuracy: 0.8700\n",
      "Epoch 13/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2707 - accuracy: 0.8800\n",
      "Epoch 14/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2456 - accuracy: 0.8700\n",
      "Epoch 15/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2489 - accuracy: 0.8800\n",
      "Epoch 16/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2432 - accuracy: 0.9000\n",
      "Epoch 17/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2646 - accuracy: 0.8800\n",
      "Epoch 18/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2314 - accuracy: 0.8900\n",
      "Epoch 19/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2615 - accuracy: 0.8600\n",
      "Epoch 20/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2483 - accuracy: 0.8800\n",
      "Epoch 21/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2465 - accuracy: 0.8600\n",
      "Epoch 22/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2605 - accuracy: 0.8700\n",
      "Epoch 23/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2445 - accuracy: 0.8800\n",
      "Epoch 24/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2384 - accuracy: 0.9000\n",
      "Epoch 25/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2531 - accuracy: 0.8800\n",
      "Epoch 26/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2399 - accuracy: 0.8700\n",
      "Epoch 27/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2289 - accuracy: 0.8900\n",
      "Epoch 28/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2367 - accuracy: 0.8700\n",
      "Epoch 29/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2400 - accuracy: 0.8800\n",
      "Epoch 30/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2360 - accuracy: 0.8900\n",
      "Epoch 31/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2390 - accuracy: 0.8900\n",
      "Epoch 32/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2304 - accuracy: 0.8900\n",
      "Epoch 33/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2360 - accuracy: 0.9000\n",
      "Epoch 34/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2275 - accuracy: 0.9000\n",
      "Epoch 35/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2433 - accuracy: 0.8800\n",
      "Epoch 36/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2611 - accuracy: 0.8800\n",
      "Epoch 37/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2477 - accuracy: 0.8700\n",
      "Epoch 38/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2810 - accuracy: 0.8800\n",
      "Epoch 39/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2835 - accuracy: 0.8700\n",
      "Epoch 40/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2296 - accuracy: 0.8900\n",
      "Epoch 41/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2863 - accuracy: 0.8800\n",
      "Epoch 42/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2476 - accuracy: 0.8500\n",
      "Epoch 43/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2460 - accuracy: 0.8900\n",
      "Epoch 44/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2781 - accuracy: 0.8800\n",
      "Epoch 45/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2535 - accuracy: 0.8600\n",
      "Epoch 46/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2534 - accuracy: 0.8700\n",
      "Epoch 47/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2886 - accuracy: 0.8600\n",
      "Epoch 48/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2699 - accuracy: 0.8700\n",
      "Epoch 49/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2461 - accuracy: 0.8700\n",
      "Epoch 50/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2457 - accuracy: 0.8900\n",
      "Epoch 51/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2486 - accuracy: 0.8700\n",
      "Epoch 52/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2288 - accuracy: 0.8900\n",
      "Epoch 53/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2490 - accuracy: 0.8900\n",
      "Epoch 54/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.3168 - accuracy: 0.8600\n",
      "Epoch 55/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.3126 - accuracy: 0.8600\n",
      "Epoch 56/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2291 - accuracy: 0.8800\n",
      "Epoch 57/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2444 - accuracy: 0.8700\n",
      "Epoch 58/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2603 - accuracy: 0.8800\n",
      "Epoch 59/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2410 - accuracy: 0.8700\n",
      "Epoch 60/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2307 - accuracy: 0.9000\n",
      "Epoch 61/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2399 - accuracy: 0.8600\n",
      "Epoch 62/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2413 - accuracy: 0.8700\n",
      "Epoch 63/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2303 - accuracy: 0.9000\n",
      "Epoch 64/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2307 - accuracy: 0.8900\n",
      "Epoch 65/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2363 - accuracy: 0.9000\n",
      "Epoch 66/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2356 - accuracy: 0.8800\n",
      "Epoch 67/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2322 - accuracy: 0.8900\n",
      "Epoch 68/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2352 - accuracy: 0.8800\n",
      "Epoch 69/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2500 - accuracy: 0.8700\n",
      "Epoch 70/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2476 - accuracy: 0.8700\n",
      "Epoch 71/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2323 - accuracy: 0.8800\n",
      "Epoch 72/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2254 - accuracy: 0.8800\n",
      "Epoch 73/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2323 - accuracy: 0.8800\n",
      "Epoch 74/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2197 - accuracy: 0.9000\n",
      "Epoch 75/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2316 - accuracy: 0.8600\n",
      "Epoch 76/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2582 - accuracy: 0.8600\n",
      "Epoch 77/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2350 - accuracy: 0.8700\n",
      "Epoch 78/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2248 - accuracy: 0.8800\n",
      "Epoch 79/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2372 - accuracy: 0.8700\n",
      "Epoch 80/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2327 - accuracy: 0.8800\n",
      "Epoch 81/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2261 - accuracy: 0.8900\n",
      "Epoch 82/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2220 - accuracy: 0.9000\n",
      "Epoch 83/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2206 - accuracy: 0.9000\n",
      "Epoch 84/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2236 - accuracy: 0.8800\n",
      "Epoch 85/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2235 - accuracy: 0.8700\n",
      "Epoch 86/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2307 - accuracy: 0.8700\n",
      "Epoch 87/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2230 - accuracy: 0.8700\n",
      "Epoch 88/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2375 - accuracy: 0.8800\n",
      "Epoch 89/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2697 - accuracy: 0.8800\n",
      "Epoch 90/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2608 - accuracy: 0.8800\n",
      "Epoch 91/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2312 - accuracy: 0.8800\n",
      "Epoch 92/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2264 - accuracy: 0.8700\n",
      "Epoch 93/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2268 - accuracy: 0.8700\n",
      "Epoch 94/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2230 - accuracy: 0.8700\n",
      "Epoch 95/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2221 - accuracy: 0.8900\n",
      "Epoch 96/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2198 - accuracy: 0.8800\n",
      "Epoch 97/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2116 - accuracy: 0.9000\n",
      "Epoch 98/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2178 - accuracy: 0.8900\n",
      "Epoch 99/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2164 - accuracy: 0.8700\n",
      "Epoch 100/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2282 - accuracy: 0.8800\n",
      "Epoch 101/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2147 - accuracy: 0.8800\n",
      "Epoch 102/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2154 - accuracy: 0.8900\n",
      "Epoch 103/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2191 - accuracy: 0.9000\n",
      "Epoch 104/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.1973 - accuracy: 0.9200\n",
      "Epoch 105/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2199 - accuracy: 0.9100\n",
      "Epoch 106/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.1894 - accuracy: 0.9300\n",
      "Epoch 107/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.2311 - accuracy: 0.8800\n",
      "Epoch 108/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2269 - accuracy: 0.9000\n",
      "Epoch 109/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.2023 - accuracy: 0.8900\n",
      "Epoch 110/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.1901 - accuracy: 0.8900\n",
      "Epoch 111/200\n",
      "4/4 [==============================] - 0s 5ms/step - loss: 0.1882 - accuracy: 0.8900\n",
      "Epoch 112/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.1939 - accuracy: 0.8700\n",
      "Epoch 113/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.1815 - accuracy: 0.8900\n",
      "Epoch 114/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.1793 - accuracy: 0.8800\n",
      "Epoch 115/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.1694 - accuracy: 0.9200\n",
      "Epoch 116/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.1612 - accuracy: 0.9500\n",
      "Epoch 117/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.1500 - accuracy: 0.9600\n",
      "Epoch 118/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.1372 - accuracy: 0.9600\n",
      "Epoch 119/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.1298 - accuracy: 0.9500\n",
      "Epoch 120/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.1197 - accuracy: 0.9600\n",
      "Epoch 121/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.1081 - accuracy: 0.9800\n",
      "Epoch 122/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.1000 - accuracy: 0.9800\n",
      "Epoch 123/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0916 - accuracy: 0.9800\n",
      "Epoch 124/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0817 - accuracy: 0.9800\n",
      "Epoch 125/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0723 - accuracy: 0.9800\n",
      "Epoch 126/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0712 - accuracy: 0.9800\n",
      "Epoch 127/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0611 - accuracy: 1.0000\n",
      "Epoch 128/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0533 - accuracy: 0.9800\n",
      "Epoch 129/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0518 - accuracy: 0.9800\n",
      "Epoch 130/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0439 - accuracy: 0.9900\n",
      "Epoch 131/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0393 - accuracy: 1.0000\n",
      "Epoch 132/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0369 - accuracy: 1.0000\n",
      "Epoch 133/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0363 - accuracy: 1.0000\n",
      "Epoch 134/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0337 - accuracy: 1.0000\n",
      "Epoch 135/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0275 - accuracy: 1.0000\n",
      "Epoch 136/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0295 - accuracy: 1.0000\n",
      "Epoch 137/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0328 - accuracy: 1.0000\n",
      "Epoch 138/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0236 - accuracy: 1.0000\n",
      "Epoch 139/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0201 - accuracy: 1.0000\n",
      "Epoch 140/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0211 - accuracy: 1.0000\n",
      "Epoch 141/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0199 - accuracy: 1.0000\n",
      "Epoch 142/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0161 - accuracy: 1.0000\n",
      "Epoch 143/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0156 - accuracy: 1.0000\n",
      "Epoch 144/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0151 - accuracy: 1.0000\n",
      "Epoch 145/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0145 - accuracy: 1.0000\n",
      "Epoch 146/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0137 - accuracy: 1.0000\n",
      "Epoch 147/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0130 - accuracy: 1.0000\n",
      "Epoch 148/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0125 - accuracy: 1.0000\n",
      "Epoch 149/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0120 - accuracy: 1.0000\n",
      "Epoch 150/200\n",
      "4/4 [==============================] - 0s 3ms/step - loss: 0.0116 - accuracy: 1.0000\n",
      "Epoch 151/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0111 - accuracy: 1.0000\n",
      "Epoch 152/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0107 - accuracy: 1.0000\n",
      "Epoch 153/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0106 - accuracy: 1.0000\n",
      "Epoch 154/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0102 - accuracy: 1.0000\n",
      "Epoch 155/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0104 - accuracy: 1.0000\n",
      "Epoch 156/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0102 - accuracy: 1.0000\n",
      "Epoch 157/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0095 - accuracy: 1.0000\n",
      "Epoch 158/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0091 - accuracy: 1.0000\n",
      "Epoch 159/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0090 - accuracy: 1.0000\n",
      "Epoch 160/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0087 - accuracy: 1.0000\n",
      "Epoch 161/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0083 - accuracy: 1.0000\n",
      "Epoch 162/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0081 - accuracy: 1.0000\n",
      "Epoch 163/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0080 - accuracy: 1.0000\n",
      "Epoch 164/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0078 - accuracy: 1.0000\n",
      "Epoch 165/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0075 - accuracy: 1.0000\n",
      "Epoch 166/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0074 - accuracy: 1.0000\n",
      "Epoch 167/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0071 - accuracy: 1.0000\n",
      "Epoch 168/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0070 - accuracy: 1.0000\n",
      "Epoch 169/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0068 - accuracy: 1.0000\n",
      "Epoch 170/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0066 - accuracy: 1.0000\n",
      "Epoch 171/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0064 - accuracy: 1.0000\n",
      "Epoch 172/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0073 - accuracy: 1.0000\n",
      "Epoch 173/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0083 - accuracy: 1.0000\n",
      "Epoch 174/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0132 - accuracy: 1.0000\n",
      "Epoch 175/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0099 - accuracy: 1.0000\n",
      "Epoch 176/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0066 - accuracy: 1.0000\n",
      "Epoch 177/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0058 - accuracy: 1.0000\n",
      "Epoch 178/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0055 - accuracy: 1.0000\n",
      "Epoch 179/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0055 - accuracy: 1.0000\n",
      "Epoch 180/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0053 - accuracy: 1.0000\n",
      "Epoch 181/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0051 - accuracy: 1.0000\n",
      "Epoch 182/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0051 - accuracy: 1.0000\n",
      "Epoch 183/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0049 - accuracy: 1.0000\n",
      "Epoch 184/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0047 - accuracy: 1.0000\n",
      "Epoch 185/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0046 - accuracy: 1.0000\n",
      "Epoch 186/200\n",
      "4/4 [==============================] - 0s 5ms/step - loss: 0.0045 - accuracy: 1.0000\n",
      "Epoch 187/200\n",
      "4/4 [==============================] - 0s 5ms/step - loss: 0.0044 - accuracy: 1.0000\n",
      "Epoch 188/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0043 - accuracy: 1.0000\n",
      "Epoch 189/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0043 - accuracy: 1.0000\n",
      "Epoch 190/200\n",
      "4/4 [==============================] - 0s 5ms/step - loss: 0.0042 - accuracy: 1.0000\n",
      "Epoch 191/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0041 - accuracy: 1.0000\n",
      "Epoch 192/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0042 - accuracy: 1.0000\n",
      "Epoch 193/200\n",
      "4/4 [==============================] - 0s 5ms/step - loss: 0.0042 - accuracy: 1.0000\n",
      "Epoch 194/200\n",
      "4/4 [==============================] - 0s 5ms/step - loss: 0.0040 - accuracy: 1.0000\n",
      "Epoch 195/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0039 - accuracy: 1.0000\n",
      "Epoch 196/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0038 - accuracy: 1.0000\n",
      "Epoch 197/200\n",
      "4/4 [==============================] - 0s 4ms/step - loss: 0.0037 - accuracy: 1.0000\n",
      "Epoch 198/200\n",
      "4/4 [==============================] - 0s 5ms/step - loss: 0.0037 - accuracy: 1.0000\n",
      "Epoch 199/200\n",
      "4/4 [==============================] - 0s 6ms/step - loss: 0.0037 - accuracy: 1.0000\n",
      "Epoch 200/200\n",
      "4/4 [==============================] - 0s 6ms/step - loss: 0.0035 - accuracy: 1.0000\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "def torch_classification():\n",
    "    \"\"\" Classification test \"\"\"\n",
    "    import matplotlib.pyplot as plt\n",
    "    from sklearn.datasets import make_moons\n",
    "    import torch\n",
    "\n",
    "    np.random.seed(12)  # 12th man, go Seahawks!\n",
    "    torch.manual_seed(12)\n",
    "\n",
    "    # Build the model\n",
    "    model = torch.nn.Sequential(\n",
    "        torch.nn.Linear(2, 10),\n",
    "        torch.nn.ReLU(),\n",
    "        torch.nn.Linear(10, 1),\n",
    "        torch.nn.Sigmoid())\n",
    "    loss = torch.nn.functional.binary_cross_entropy\n",
    "    opt = torch.optim.Adam(model.parameters(), lr=.1)\n",
    "\n",
    "    # Train it\n",
    "    x, y = make_moons(n_samples=100, noise=0.1)     # Two interlevered half circles\n",
    "    y = y.reshape ((100, 1))                        # Make y an array of 1x1 arrays to match the output shape of the model\n",
    "    y = torch.Tensor(y)\n",
    "    for epoch in range(200):\n",
    "        yhat = model(torch.Tensor(x))\n",
    "        l = loss(yhat, torch.Tensor(y))\n",
    "        opt.zero_grad()\n",
    "        l.backward()\n",
    "        opt.step()\n",
    "        acc = (len(y) - torch.logical_xor(yhat > .5, y > .5).sum()) / len(y)\n",
    "        print (f\"Epoch {epoch}: loss={l.item():.4f}, accuracy={acc * 100:.0f}%\")\n",
    "\n",
    "    # visualize the decision boundary\n",
    "    x1 = np.linspace(-2, 2.5, 100)\n",
    "    x2 = np.linspace(-2, 2, 100)\n",
    "    xx1, xx2 = np.meshgrid(x1, x2)\n",
    "    X = np.hstack((xx1.reshape(-1,1), xx2.reshape(-1,1)))\n",
    "    yhat = model(torch.Tensor(X)).detach().numpy()\n",
    "    plt.contourf(x1, x2, np.round(yhat.reshape(100,100)), cmap='jet', alpha=.5)\n",
    "    plt.scatter(x[:,0], x[:,1], c=y, s=20, cmap='jet')\n",
    "    plt.title(\"PyTorch Classification\")\n",
    "    plt.show()\n",
    "\n",
    "def tf_classification():\n",
    "    import keras\n",
    "    import  numpy as np\n",
    "    import  matplotlib.pyplot as plt\n",
    "    from    sklearn.datasets import make_moons\n",
    "\n",
    "    # Build the model\n",
    "    model = keras.models.Sequential([keras.layers.Dense(10, input_dim=2, activation=\"relu\"),\n",
    "                       keras.layers.Dense(1, activation=\"sigmoid\")])\n",
    "    loss = keras.losses.BinaryCrossentropy()\n",
    "    opt = keras.optimizers.Adam(learning_rate=0.1)\n",
    "    model.compile(loss=loss, optimizer=opt, metrics=['accuracy'])\n",
    "\n",
    "    # Train it\n",
    "    x, y = make_moons(n_samples=100, noise=0.1)     # Two interlevered half circles\n",
    "    y = y.reshape ((100, 1))                        # Make y an array of 1x1 arrays to match the output shape of the model\n",
    "    model.fit(x, y, epochs=200)\n",
    "\n",
    "    # Visualize the decision boundary\n",
    "    x1 = np.linspace(-2, 2.5, 100)\n",
    "    x2 = np.linspace(-2, 2, 100)\n",
    "    xx1, xx2 = np.meshgrid(x1, x2)\n",
    "    X = np.hstack((xx1.reshape(-1,1), xx2.reshape(-1,1)))\n",
    "    yhat = model(X).numpy()\n",
    "    plt.contourf(x1, x2, np.round(yhat.reshape(100,100)), cmap='jet', alpha=.5)\n",
    "    plt.scatter(x[:,0], x[:,1], c=y, s=20, cmap='jet')\n",
    "    plt.title(\"TensorFlow Classification\")\n",
    "    plt.show()\n",
    "\n",
    "torch_classification()\n",
    "tf_classification()\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "base",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.10"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}