X *= srng.binomial(X.shape, p=retain_prob, dtype=theano.config.floatX)
X /= retain_prob
return X
def model(X, w_h, w_h2, w_o, p_drop_input, p_drop_hidden):
X = dropout(X, p_drop_input)
h = rectify(T.dot(X, w_h))
h = dropout(h, p_drop_hidden)
h2 = rectify(T.dot(h, w_h2))
h2 = dropout(h2, p_drop_hidden)
py_x = softmax(T.dot(h2, w_o))
return h, h2, py_x
col, trX, trY=load.geotutor()
col, trX, trY=load.process(col, trX, trY)
X = T.fmatrix()
Y = T.fmatrix()
w_h = init_weights((364, 59))
w_h2 = init_weights((59, 59))
w_o = init_weights((59, 2))
noise_h, noise_h2, noise_py_x = model(X, w_h, w_h2, w_o, 0.2, 0.5)
h, h2, py_x = model(X, w_h, w_h2, w_o, 0., 0.)
y_x = T.argmax(py_x, axis=1)
params = [w_h, w_h2, w_o]
#cost=T.mean(T.nnet.categorical_crossentropy(noise_py_x, Y))+ 0.01 * ((w_h ** 2).sum()+(w_h2 ** 2).sum()+(w_o ** 2).sum())
trueY=trY[:num]
predY=predict(trX[:num])
train_acc=np.mean(np.argmax(trY[:num], axis=1) == predict(trX[:num]))
#print num, precision_recall_fscore_support(list(y[1] for y in trueY.tolist()),predY.tolist(), average='micro')
print num, train_acc,test_acc
return train_acc,test_acc
def data_split(data,targets):
trainX=data[:4908]
trainY=targets[:4908]
testX=data[4908:]
testY=targets[4908:]
return trainX,testX,trainY,testY
col, data, targets=load.geotutor()
col, data, targets=load.process(col, data, targets)
#trX,teX,trY,teY=cross_validation.train_test_split(data,targets,test_size=0.3,random_state=0)
trX,teX,trY,teY=data_split(data,targets)
X=T.fmatrix()
Y=T.fmatrix()
w=init_weights((407, 2))
py_x=model(X,w)
y_pred=T.argmax(py_x, axis=1)
cost=T.mean(T.nnet.categorical_crossentropy(py_x, Y))+0.001*T.sum(w**2)
gradient=T.grad(cost=cost, wrt=w)
update=[[w,w-gradient*0.005]]
