Solutions and bugfixes for part1

part1/2023-CoDaS-HEP-Exercises-1.ipynb

@@ -168,9 +168,9 @@
    "outputs": [],
    "source": [
     "F = np.array([0, 12, 45.21, 34, 99.91])\n",
-    "C = np.zeros(5) # Your solution here\n",
-    "\n",
-    "C_exp = np.array([-17.77777778,  42.22222222, 208.27222222, 152.22222222, 481.77222222])\n",
+    "C = (F - 32) * 5 / 9\n",
+    "print(C)\n",
+    "C_exp = np.array([-17.77777778, -11.11111111, 7.33888889, 1.11111111, 37.72777778])\n",
     "\n",
     "assert np.all(np.isclose(C_exp, C, rtol=1e-05, atol=1e-08, equal_nan=False))"
    ]
@@ -378,44 +378,47 @@
     "    def __init__(self):\n",
     "        self.coef_ = None\n",
     "        self.intercept_ = None\n",
-    "        self.init_weights()\n",
     "\n",
     "    def init_weights(self, num_features):\n",
     "        \"\"\"Accepts an input num_features and sets the coef_\n",
     "        and intercept_ attributes to random values drawn from a normal distribution\n",
     "        with mean 0 and standard deviation 1.\n",
     "        Use a numpy array for coef_ and a float for intercept_\"\"\"\n",
     "        # Hint: use `np.random.normal`\n",
-    "        self.coef_ = None # TODO: Change\n",
-    "        self.intercept_ = None # TODO: Change\n",
+    "        self.coef_ = np.random.normal(size=num_features)\n",
+    "        self.intercept_ = np.random.normal()\n",
     "        \n",
     "    def score(self, X, y_true):\n",
     "        \"\"\"Accepts inputs X and y_true and outputs the r2 score of the model.\"\"\"\n",
     "        y_pred = self.predict(X)\n",
-    "        pass\n",
+    "        u = ((y_true - y_pred)**2).sum()\n",
+    "        v = ((y_true - y_true.mean())**2).sum()\n",
+    "        return 1 - u/v\n",
     "\n",
     "    def calc_loss(self, X, y_true):\n",
     "        \"\"\"Calculates the loss value using current coef/intercept values\"\"\"\n",
     "        y_pred = self.predict(X)\n",
-    "        pass\n",
+    "        return np.mean((y_pred - y_true)**2)/2.\n",
     "\n",
     "    def predict(self, X):\n",
     "        \"\"\"Creates a prediction with the current coef/intercept values\"\"\"\n",
-    "        pass\n",
+    "        return ([email protected]_ + self.intercept_)\n",
     "\n",
     "    def fit(self, X, y_true, max_iter=10000, learning_rate=0.01):\n",
     "        \"\"\"Optimizing models according to a gradient as a way of minimizing loss.\n",
     "        Should update the coef_ and intercept_ attributes\"\"\"\n",
     "        # Hint: https://ml-cheatsheet.readthedocs.io/en/latest/gradient_descent.html#step-by-step\n",
     "        for _ in range(max_iter):\n",
     "            coef_grad, intercept_grad = self.calc_grad(X, y_true)\n",
-    "            self.coef_ -= None # TODO: Change\n",
-    "            self.intercept_ -= None # TODO: Change\n",
+    "            self.coef_ -= learning_rate * coef_grad\n",
+    "            self.intercept_ -= learning_rate * intercept_grad\n",
     "    \n",
     "    def calc_grad(self, X, y_true):\n",
     "        \"\"\"Calculates gradients for coef/intercept values\"\"\"\n",
-    "        coef_grad = None # TODO: Change\n",
-    "        intercept_grad = None # TODO: Change\n",
+    "        res = y_true - self.predict(X)\n",
+    "        res_mat = np.tile(res.reshape(-1, 1), (1, X.shape[1]))\n",
+    "        coef_grad = -1 * np.mean(res_mat * X, axis=0)\n",
+    "        intercept_grad = -1 * np.mean(res)\n",
     "        return coef_grad, intercept_grad"
    ]
   },
@@ -453,7 +456,7 @@
    "source": [
     "import matplotlib.pyplot as plt\n",
     "\n",
-    "plt.scatter(target, lr.predict(inputs))\n",
+    "plt.scatter(target, my_regressor.predict(inputs))\n",
     "plt.plot([0, 100], [0, 100], 'r')\n",
     "plt.xlabel('Expected')\n",
     "plt.ylabel('Predicted');"