def test_balanced_categorical_crossentropy():
samples = (2, )
img_dim = (3, 3)
num_classes = 5
y_true_n = np.random.randint(0, num_classes, samples + img_dim)
y_true_n = dv.one_hot(y_true_n, num_classes)
y_true = K.variable(y_true_n)
y_pred_n = np.random.random(samples + img_dim + (num_classes, )).astype(
K.floatx())
y_pred = K.variable(y_pred_n)
y_pred = softmax(y_pred)
loss = dv.tf.balanced_categorical_crossentropy(
y_true, y_pred).eval(session=K.get_session())
print(loss)
def out_adapt(x, relabel_LVOT, target=adapt_size, n=cg.num_classes):
return dv.one_hot(out_adapt_raw(x, relabel_LVOT, target), n)
def out_adapt(self, x, target=cg.dim):
return dv.one_hot(self.out_adapt_raw(x, target), cg.num_classes)