def reduceMNIST(n):
Xtrain, Ytrain = reader.readmnistAsOneLineTraining('mnist\\train.csv')
Xtest = reader.readmnistAsOneLineTest('mnist\\test.csv')
from sklearn.decomposition import PCA
pca = PCA(n_components=784)
Xtrain = np.array(Xtrain).reshape((42000, 784))
pca = pca.fit(Xtrain)
cum_var_explained = np.cumsum(pca.explained_variance_) / np.sum(
pca.explained_variance_)
plt.figure(10000)
plt.title("Wykres wytłumaczonej wariancji zbioru")
plt.ylabel("Część wytłumaczonej wariancji")
plt.xlabel("Liczba wymiarów")
plt.grid(True)
plt.plot(cum_var_explained)
plt.show()
def reduceMNIST(n):
Xtrain, Ytrain = reader.readmnistAsOneLineTraining('mnist/train.csv')
Xtest = reader.readmnistAsOneLineTest('mnist/test.csv')
from sklearn.decomposition import PCA
# pca = PCA(n_components = 784)
# Xtrain = np.array(Xtrain).reshape((42000, 784))
# pca = pca.fit(Xtrain)
# cum_var_explained = np.cumsum(pca.explained_variance_)
# plt.figure(10000)
# plt.plot(cum_var_explained)
# plt.show()
pca = PCA(n_components = n)
Xtrain = pca.fit_transform(np.array(Xtrain).reshape((42000, 784)))
np.savetxt('mnist/trainPCA'+str(n)+'.csv', np.round(Xtrain, 5), delimiter=',')
Xtest = np.array(Xtest).reshape((28000, 784))
Xtest = np.round(pca.transform(Xtest), 5)
np.savetxt('mnist/testPCA'+str(n)+'.csv', Xtest, delimiter=',')
def test0():
name = 'test0'
Xtrain, Ytrain = reader.readmnistAsOneLineTraining('mnist/train.csv')
print('Training set ready, size: ', len(Xtrain))
net = nn.NeuralNetwork(momentumSize=0)
net.addLayer(784, 10, F.softmax, True)
net.setCostFunction(L.crossEntropy)
print('Starting training...')
Xtest = reader.readmnistAsOneLineTest('mnist/test.csv')
net.kFoldsTrainAndValidate(Xtrain, Ytrain,
k=6,
epochs=100,
learningRate=1e-2,
batchSize=100,
showError=True,
showNodes=False,
print= lambda e: _test(e, Xtest, net, name),
showEvery=5,
name = name)
def test3():
_, Ytrain = reader.readmnistAsOneLineTraining('mnist\\train.csv')
Xtrain = reader.readmnistAsOneLinePCA('mnist\\trainPCA300.csv')
Xtest = reader.readmnistAsOneLinePCA('mnist\\testPCA300.csv')
name = '3test'
print('Training set ready, size: ', len(Xtrain))
net = nn.NeuralNetwork(momentumSize=0)
net.addLayer(300, 150, F.tanh, True)
net.addLayer(150, 70, F.LReLU, True)
net.addLayer(70, 10, F.softmax, True)
net.setCostFunction(L.crossEntropy)
print('Starting training...')
net.kFoldsTrainAndValidate(Xtrain,
Ytrain,
k=6,
epochs=100,
learningRate=1e-2,
batchSize=100,
showError=True,
showNodes=False,
print=lambda e: _test(e, Xtest, net, name),
showEvery=25,
name=name)