def save_test_summary(x_batch, y_batch, x_strings_batch, name_):
'''
save info for a batch in order to plot in
bokeh later
'''
path_ = os.path.join(out_dir, name_)
y_net = []
prob_net = []
layer = []
true_labels = []
feed_dict = {
x_values: x_batch,
y_values: y_batch,
dropout_prob: 1.0
}
output_ = [graph_predictions, graph_true_predictions,
graph_probs, graph_state_]
predictions, true_pred, probs, fc_layer = sess.run(
output_, feed_dict)
prob_net = probs.tolist()
layer = fc_layer.tolist()
y_net = predictions.tolist()
true_labels = true_pred.tolist()
process_utils.save_info(
x_strings_batch, true_labels, y_net, prob_net, layer, path_)
def save_dev_summary(x_batch, y_batch, x_strings_batch, name_):
'''
save info for a batch in order to plot in
bokeh later
'''
path_ = os.path.join(sent_dir, name_)
y_net = []
prob_net = []
layer = []
true_labels = []
feed_dict = {
network.x: x_batch,
network.y: y_batch,
network.dropout_prob: 1.0
}
output_ = [
network.predictions, network.true_predictions, network.probs,
network.h_pool_flat
]
predictions, true_pred, probs, fc_layer = sess.run(output_, feed_dict)
prob_net = probs.tolist()
layer = fc_layer.tolist()
y_net = predictions.tolist()
true_labels = true_pred.tolist()
process_utils.save_info(x_strings_batch, true_labels, y_net, prob_net,
layer, path_)
def save_test_summary(x_batch, y_batch, x_strings_batch, name_):
'''
save info for a batch in order to plot in
bokeh later
'''
path_ = os.path.join(out_dir, name_)
y_net = []
prob_net = []
layer = []
true_labels = []
flag_ = True
if config['split_dev']:
mini_size = config['dev_minibatch']
for i in range(0, len(x_batch), mini_size):
if (i + mini_size < len(x_batch)):
mini_x_batch = x_batch[i:i + mini_size]
mini_y_batch = y_batch[i:i + mini_size]
else:
mini_x_batch = x_batch[i:]
mini_y_batch = y_batch[i:]
dropouts = [1.0, 1.0, 1.0]
reg_metrics = [1, 0, 0]
feed_dict = make_feed_dict(
mini_x_batch, mini_y_batch,
dropouts, reg_metrics, question)
output_ = [graph_predictions, graph_true_predictions,
graph_probs, graph_state_, graph_embeddings]
predictions, true_pred, probs, fc_layer, embds_ = sess.run(
output_, feed_dict)
if flag_: # save word_embeddings
flag_ = False
save_embeddings(embds_)
prob_net += probs.tolist()
layer += fc_layer.tolist()
y_net += predictions.tolist()
true_labels += true_pred.tolist()
else:
print (
"doesn't support having input as a single batch!! Set:"
"config['split_dev'] to True ")
sys.exit(1)
# print (
# len(x_strings_batch), len(true_labels), len(y_net),
# len(prob_net), len(layer))
process_utils.save_info(
x_strings_batch, true_labels, y_net, prob_net, layer, path_)
print ("saved info to: {}".format(path_))
def save_dev_summary(x_batch, y_batch, x_strings_batch, name_):
'''
save info for a batch in order to plot in
bokeh later
'''
path_ = os.path.join(out_dir, name_)
y_net = []
prob_net = []
layer = []
true_labels = []
if config['split_dev']:
mini_size = config['dev_minibatch']
for i in range(0, len(x_batch), mini_size):
if (i + mini_size < len(x_batch)):
mini_x_batch = x_batch[i:i + mini_size]
mini_y_batch = y_batch[i:i + mini_size]
else:
mini_x_batch = x_batch[i:]
mini_y_batch = y_batch[i:]
dropouts = [1.0, 1.0, 1.0]
reg_metrics = [1, 0, 0]
feed_dict = make_feed_dict(mini_x_batch, mini_y_batch,
dropouts, reg_metrics, question)
output_ = [
network.predictions, network.true_predictions,
network.probs, network.state_
]
predictions, true_pred, probs, fc_layer = sess.run(
output_, feed_dict)
# print (predictions)
# print (probs)
# print (fc_layer)
# print (fc_layer.shape)
# print (true_pred)
prob_net += probs.tolist()
layer += fc_layer.tolist()
y_net += predictions.tolist()
true_labels += true_pred.tolist()
else:
print("doesn't support having input as a single batch!! Set:"
"config['split_dev'] to True ")
sys.exit(1)
# print (
# len(x_strings_batch), len(true_labels), len(y_net),
# len(prob_net), len(layer))
process_utils.save_info(x_strings_batch, true_labels, y_net, prob_net,
layer, path_)