def test_mean_squared_derivative(backend_default):
outputs = np.array([0.5, 1.0, 0.0, 0.0001]).reshape((4, 1))
targets = np.array(([0.5, 0.0, 1.0, 0.2])).reshape((4, 1))
expected_result = (outputs - targets) / outputs.shape[1] / outputs.shape[0]
compare_tensors(MeanSquared(),
outputs,
targets,
expected_result,
deriv=True,
tol=1e-8)
def test_mean_squared_limits(backend_default):
outputs = np.array([0.5, 1.0, 0.0, 0.0001]).reshape((4, 1))
targets = np.array(([0.5, 0.0, 1.0, 0.2])).reshape((4, 1))
expected_result = np.mean((outputs - targets) ** 2, axis=0, keepdims=True) / 2.
compare_tensors(MeanSquared(), outputs, targets, expected_result, tol=1e-7)
reset_cells=False),
Affine(train_set.nfeatures, init, bias=init, activation=Identity())
]
else:
layers = [
LSTM(hidden,
init,
activation=Logistic(),
gate_activation=Tanh(),
reset_cells=True),
RecurrentLast(),
Affine(train_set.nfeatures, init, bias=init, activation=Identity())
]
model = Model(layers=layers)
cost = GeneralizedCost(MeanSquared())
optimizer = RMSProp(stochastic_round=args.rounding)
callbacks = Callbacks(model, eval_set=valid_set, **args.callback_args)
# fit model
model.fit(train_set,
optimizer=optimizer,
num_epochs=args.epochs,
cost=cost,
callbacks=callbacks)
# =======visualize how the model does on validation set==============
# run the trained model on train and valid dataset and see how the outputs match
train_output = model.get_outputs(train_set).reshape(
-1, train_set.nfeatures)
layers = [
Affine(nout=50, init=w, bias=b, activation=Rectlin()),
Dropout(keep=0.5),
Affine(nout=50, init=w, bias=b, activation=Rectlin()),
Dropout(keep=0.4),
Affine(nout=3, init=w, bias=b, activation=Softmax()),
Dropout(keep=0.3)
]
# Optimizer
optimizer = GradientDescentMomentum(0.1,
momentum_coef=0.9,
stochastic_round=args.rounding)
# Cost
cost = GeneralizedCost(costfunc=MeanSquared())
model = Model(layers=layers)
callbacks = Callbacks(model, eval_set=val_iter, **args.callback_args)
# Training
model.fit(train_iter,
optimizer=optimizer,
num_epochs=1,
cost=cost,
callbacks=callbacks)
# Evluate
evaluate(model, val_iter, Metric=Misclassification())
