def eval(model, gen, x, y):
"""Validation function.
Args:
model: keras model.
gen: object, data generator.
x: 3darray, input, (n_segs, n_concat, n_freq)
y: 2darray, target, (n_segs, n_freq)
"""
pred_all, y_all = [], []
# Inference in mini batch.
for (batch_x, batch_y) in gen.generate(xs=[x], ys=[y]):
pred = model.predict(batch_x)
pred_all.append(pred)
y_all.append(batch_y)
if False:
print("pred")
print(pred)
# Concatenate mini batch prediction.
pred_all = np.concatenate(pred_all, axis=0)
y_all = np.concatenate(y_all, axis=0)
# Compute loss.
loss = pp_data.np_mean_absolute_error(y_all, pred_all)
return loss
def eval(sess, model, gen, x, y):
"""Validation function.
Args:
model: keras model.
gen: object, data generator.
x: 3darray, input, (n_segs, n_concat, n_freq)
y: 2darray, target, (n_segs, n_freq)
"""
pred_all, y_all = [], []
# Inference in mini batch.
for (batch_x, batch_y) in gen.generate(xs=[x], ys=[y]):
pred = sess.run([model.enhanced_outputs],
feed_dict={model.x_noisy: batch_x})
pred = np.reshape(pred, (-1, batch_y.shape[1]))
pred_all.append(pred)
# batch_y = np.reshape(batch_y,(-1, batch_y.shape[1]))
y_all.append(batch_y)
# Concatenate mini batch prediction.
pred_all = np.concatenate(pred_all, axis=0)
y_all = np.concatenate(y_all, axis=0)
# Compute loss.
loss = pp_data.np_mean_absolute_error(y_all, pred_all)
return loss
def eval(model, gen, x1, x2, y1, y2, name, utts):
"""Validation function.
Args:
model: keras model.
gen: object, data generator.
x: 3darray, input, (n_segs, n_concat, n_freq)
y: 2darray, target, (n_segs, n_freq)
"""
pred_all, y_all = [], []
# Inference in mini batch.
for (batch_x, batch_y) in gen.generate([x1, x2, name], [y1, y2], utts):
pred = model.predict(batch_x)
pred_all.append(np.hstack(pred))
y_all.append(np.hstack(batch_y))
# Concatenate mini batch prediction.
pred_all = np.concatenate(pred_all, axis=0)
y_all = np.concatenate(y_all, axis=0)
# Compute loss.
loss = pp_data.np_mean_absolute_error(y_all, pred_all)
return loss