def check_loss(self, X, Y, Ws=[]):
"""Check loss at output layer for observations X/Y."""
if (len(Ws) == 0):
Ws = self.layer_weights()
obs_count = X.shape[0]
M = self.get_drop_masks(obs_count, 0, 0)
A = self.feedforward(X, M)
O = self.out_loss(A[-1], Y)
Yc = lnf.class_cats(Y)
Yhc = lnf.class_cats(A[-1])
hits = sum([(int(a) == int(b)) for (a, b) in zip(Yhc, Yc)])
acc = float(hits) / float(obs_count)
return {'loss': O['L'], 'acc': acc, 'grad': O['dL']}
def check_acc(self, X, Y, Ws=[]):
"""Check classification accuracy using inputs X with classes Y.
Classes should be represented in Y as a +/- 1 indicator matrix.
"""
if (len(Ws) == 0):
Ws = self.layer_weights()
obs_count = X.shape[0]
Yc = lnf.class_cats(Y)
M = self.get_drop_masks(obs_count, 0, 0)
A = self.feedforward(X, M, Ws)
Yh = A[-1]
Yhc = lnf.class_cats(Yh)
hits = sum([(int(a) == int(b)) for (a, b) in zip(Yhc, Yc)])
return (float(hits) / float(obs_count))
