output_size = 8
N = 120 # number of memory locations
M = 8 # size of a memory location
# model initialization
layers = [
GRU(hidden_size, init, activation=Tanh(), gate_activation=Logistic()),
Affine(train_set.nout, init, bias=init, activation=Logistic())
]
cost = GeneralizedCostMask(costfunc=CrossEntropyBinary())
model = Model(layers=layers)
optimizer = RMSProp(gradient_clip_value=gradient_clip_value,
stochastic_round=args.rounding)
# configure callbacks
callbacks = Callbacks(model, **args.callback_args)
# we can use the training set as the validation set,
# since the data is tickerally generated
callbacks.add_watch_ticker_callback(train_set)
# train model
model.fit(train_set,
optimizer=optimizer,
num_epochs=args.epochs,
cost=cost,
callbacks=callbacks)
