In [1]:

    

import datasets.cifar10 as cifar10
import numpy as np

mnist = np.loadtxt('datasets/mnist/train.csv', skiprows=1, delimiter=',')

trainX = mnist[:32000, 1:]
trainY = mnist[:32000, 0]
testX = mnist[32000:, 1:]
testY = mnist[32000:, 0]

oneHotTrainY = np.zeros(shape=(len(np.unique(trainY)), trainX.shape[0]))
for i in range(0, trainX.shape[0]):
    oneHotTrainY[trainY[i], i] = 1
mean, std = np.mean(trainX), np.std(trainX)

trainX = (trainX - mean)/std
testX = (testX - mean)/std

oneHotTestY = np.zeros(shape=(len(np.unique(trainY)), testX.shape[0]))
for i in range(0, testX.shape[0]):
    oneHotTestY[testY[i], i] = 1


    

    




/Users/kirill/anaconda/lib/python3.5/site-packages/ipykernel/__main__.py:13: VisibleDeprecationWarning: using a non-integer number instead of an integer will result in an error in the future
/Users/kirill/anaconda/lib/python3.5/site-packages/ipykernel/__main__.py:21: VisibleDeprecationWarning: using a non-integer number instead of an integer will result in an error in the future

In [5]:

    

import NeuralNetwork2Layer

nn = NeuralNetwork2Layer.NeuralNetworkClassifier(50, 10, epochs=700, learningRate=0.01, batchSize=256, l2=0.001)
nn.fit(trainX, oneHotTrainY, validationFunc = lambda model: np.sum(model.predict(testX) == testY)/len(testX))


    

    




0%                                                        100%
[████████████████████████████████████████████████████████████] | ETA: 00:00:00 | Item ID: [0.997, 0.974] 0.10, 0.09
Total time elapsed: 00:08:16
Title: 
  Started: 07/13/2016 23:08:45
  Finished: 07/13/2016 23:17:02
  Total time elapsed: 00:08:16

In [ ]:

    

import matplotlib.pyplot as plt
%matplotlib inline

fig, ax = plt.subplots(nrows=2, ncols=2)
ax[0, 0].scatter(np.arange(0, len(nn.batchLosses)), nn.batchLosses)
plt.xlabel('Batch Epoch')
plt.ylabel('Batch Loss')

ax[0, 1].scatter(np.arange(0, len(nn.losses)), nn.losses)
plt.xlabel('Epoch')
plt.ylabel('Loss')


ax[1, 0].scatter(np.arange(0, len(nn.trainAccuracies)), nn.trainAccuracies, color='g')
ax[1, 0].axhline(1)
plt.xlabel('Epoch')
plt.ylabel('Accuracy')

fig.set_size_inches((20, 15))