# cost function
cost = (nll_multiclass(mlp.output, it) + L1_reg * mlp.L1 + L2_reg * mlp.L2_sqr)
pred = pred_multiclass(mlp.output)
errors = pred_error(pred, it)
params = flatten(mlp.params)
print "training the MLP with rmsprop"
optimize(dataset=dataset,
inputs=inputs,
cost=cost,
params=params,
errors=errors,
n_epochs=1000,
batch_size=20,
patience=5000,
patience_increase=1.5,
improvement_threshold=0.995,
optimizer="rmsprop")
print "compiling the prediction function"
predict = theano.function(inputs=[x], outputs=pred)
distribution = theano.function(inputs=[x], outputs=mlp.output)
print "predicting the first 10 samples of the test dataset"
print "predict:", predict(mnist[2][0][0:10])
print "answer: ", mnist[2][1][0:10]
print "with dropout, the output distributions should all be slightly different"
nll_multiclass(mlp.output, it)
+ L1_reg * mlp.L1
+ L2_reg * mlp.L2_sqr
)
pred = pred_multiclass(mlp.output)
errors = pred_error(pred, it)
params = flatten(mlp.params)
print "training the MLP with adadelta"
optimize(dataset=dataset,
inputs=inputs,
cost=cost,
params=params,
errors=errors,
n_epochs=1000,
batch_size=20,
patience=5000,
patience_increase=1.5,
improvement_threshold=0.995,
optimizer="adadelta")
print ""
print "compiling the prediction function"
predict = theano.function(inputs=[x], outputs=pred)
print "predicting the first 10 samples of the test dataset"
print "predict:", predict(mnist[2][0][0:10])
print "answer: ", mnist[2][1][0:10]
pred = pred_multiclass(mlp.output)
errors = pred_error(pred, it)
params = flatten(mlp.params)
print "training the MLP with rmsprop"
losses = optimize(
dataset=dataset,
inputs=inputs,
cost=cost,
params=params,
errors=errors,
n_epochs=2,
batch_size=20,
patience=5000,
patience_increase=1.5,
improvement_threshold=0.995,
optimizer='rmsprop'
)
print "compiling the prediction function"
predict = theano.function(inputs=[x], outputs=pred)
print "predicting the first 10 samples of the test dataset"
print "predict:", predict(mnist[2][0][0:10])
print "answer: ", mnist[2][1][0:10]
)
pred = pred_multiclass(y)
errors = pred_error(pred, it)
params = flatten(layers_params(layers))
print "training the LSTM with adadelta"
optimize(dataset=dataset,
inputs=inputs,
cost=cost,
params=params,
errors=errors,
n_epochs=200,
batch_size=64,
patience=1500,
patience_increase=1.25,
improvement_threshold=0.995,
test_batches=1,
print_cost=True,
optimizer="adadelta")
print "compiling the prediction function"
predict = theano.function(inputs=[x, mask], outputs=pred)
print "predicting the first 10 samples of the test dataset"
print "predict:", predict(dataset[2][0].get_value()[0:10], dataset[2][-1].get_value()[0:10])
print "answer: ", dataset[2][1].get_value()[0:10]
+ L1_reg * mlp.L1
+ L2_reg * mlp.L2_sqr
)
pred = pred_multiclass(mlp.output)
errors = pred_error(pred, it)
params = flatten(mlp.params)
print "training the MLP with sgd"
optimize(dataset=dataset,
inputs=inputs,
cost=cost,
params=params,
errors=errors,
learning_rate=0.01,
momentum=0.2,
n_epochs=1000,
batch_size=20,
patience=1000,
patience_increase=1.5,
improvement_threshold=0.995,
optimizer="sgd")
print "compiling the prediction function"
predict = theano.function(inputs=[x], outputs=pred)
print "predicting the first 10 samples of the test dataset"
print "predict:", predict(mnist[2][0][0:10])
print "answer: ", mnist[2][1][0:10]