A simple implementation of an autograd framework for javascript with a Multi Layer Perceptron that trains against a toy dataset. Based on the lesson from Andrej Karpathy.
Browser console should output something similar to the following:
step 0: 1.576735949332759
step 10: 0.738121537851598
step 20: 0.22560775316053272
step 30: 0.09944237732767153
step 40: 0.06201518984620813
step 50: 0.04451835351766176
step 60: 0.03445270000447273
step 70: 0.027945728570807543
step 80: 0.023411597703849537
step 90: 0.02008185196292793