print('epoch %d end at %s' % (epoch, str(end)))
print('epoch %d train loss: %f' % (epoch, train_loss))
print('epoch %d dev loss: %f' % (epoch, dev_loss))
print('epoch %d dev accuracy: %f' % (epoch, epcoh_accuracy))
print('best epoch now: %d\n' % best_epoch)
log.write('epoch %d end at %s\n' % (epoch, str(end)))
log.write('epoch %d train loss: %f\n' % (epoch, train_loss))
log.write('epoch %d dev loss: %f\n' % (epoch, dev_loss))
log.write('epoch %d dev accuracy: %f\n' % (epoch, epcoh_accuracy))
log.write('best epoch now: %d\n\n' % best_epoch)
end_time = datetime.now()
print('train end at %s\n' % str(end_time))
log.write('train end at %s\n\n' % str(end_time))
timedelta = end_time - start_time
print('train cost time: %s\n' % str(timedelta))
print('best epoch last: %d\n' % best_epoch)
log.write('train cost time: %s\n\n' % str(timedelta))
log.write('best epoch last: %d\n\n' % best_epoch)
plot.plot_loss(all_train_loss,
all_dev_loss,
folder_path=folder_path,
title='%s' % model_name)
plot.plot_accuracy(all_dev_accuracy,
folder_path=folder_path,
title='%s' % model_name)
train_y = np.load("output/train_anno.npy")
train_names = np.load("output/train_names.npy")
for name in runs[:1]:
feature_type = name.split('_')[0]
test_x = np.load("output/val_{}.npy".format(feature_type))
train_x = np.load("output/train_{}.npy".format(feature_type))
num_train, input_dim = train_x.shape
_, output_dim = train_y.shape
if TRAIN:
loss_hist = simple_nn.train(input_dim, output_dim, train_x,
train_y, test_x, test_y, name)
plot.plot_loss(loss_hist, name)
model = simple_nn.load(name)
# make predictions on the test data
train_loss = model.evaluate(train_x, train_y, verbose=0)
val_loss = model.evaluate(test_x, test_y, verbose=0)
print("{} - train: {}, val {}".format(model.metrics_names, train_loss,
val_loss))
with open('logs/nn_losses.txt', 'a') as f:
f.write('{} {} {} {}\n'.format(int(time.time()), name, train_loss,
val_loss))
'''
print("[INFO] predicting...")
preds = model.predict(test_x)
evaluate.evaluate_point_normal(test_y, preds)