def generate_new_data():
global trn_inputs, trn_targets
#print "Generating new data..."
trn_inputs, trn_targets = \
generate_id_data(cfg.x_len, cfg.n_batch)
f_output = function(inputs=[ps.flat, x], outputs=out_re)
f_loss = function(inputs=[ps.flat, x, t], outputs=loss)
f_pure_loss = function(inputs=[ps.flat, x, t], outputs=pure_loss)
f_dloss = function(inputs=[ps.flat, x, t], outputs=T.grad(loss, ps.flat))
# separate gradients wrt layer weights
if show_gradient:
f_grads = {}
for wname, wvar in ps.vars.iteritems():
f_grads[wname] = function(inputs=[ps.flat, x, t], outputs=T.grad(loss, wvar))
if do_weight_plots:
plt.figure()
print "Generating validation data..."
val_inputs, val_targets = generate_id_data(cfg.x_len, cfg.n_val_samples)
tst_inputs, tst_targets = generate_id_data(cfg.x_len, cfg.n_val_samples)
print "Done."
# optimizer
def generate_new_data():
global trn_inputs, trn_targets
#print "Generating new data..."
trn_inputs, trn_targets = \
generate_id_data(cfg.x_len, cfg.n_batch)
def f_trn_loss(p):
global trn_inputs, trn_targets
if check_nans:
assert np.all(np.isfinite(gather(p))), "NaN in p given to f_trn_loss"
return f_loss(p, trn_inputs, trn_targets)
