def on_epoch_end(self, epoch, logs={}):
self.sum += time.time() - self.epoch_time_start
if epoch < 30:
recorder_step = 5
elif epoch < 300:
recorder_step = 50
else:
recorder_step = 150
if epoch % recorder_step == 0:
self.intervals.append(self.sum)
self.sum = 0
y_true = y_test
y_pred = autoencoder.predict(x_test)
pred_index, true_index = dblp_eval.find_indices(y_pred, y_true)
self.ndcg.append(ndcg_metric(pred_index, true_index))
self.map.append(map_metric(pred_index, true_index))
true_indices = []
pred_indices = []
x_test = np.asarray([
x_test_record[0].todense() for x_test_record in x_test
]).reshape(x_test.__len__(), -1)
y_test = np.asarray([
y_test_record[0].todense() for y_test_record in y_test
]).reshape(y_test.__len__(), -1)
with open(result_output_name, 'a+') as file:
writer = csv.writer(file)
for sample_x, sample_y in zip(x_test, y_test):
start_time = time.time()
sample_prediction = autoencoder.predict(np.asmatrix(sample_x))
end_time = time.time()
elapsed_time = (end_time - start_time) * 1000
pred_index, true_index = dblp_eval.find_indices(
sample_prediction, [sample_y])
true_indices.append(true_index[0])
pred_indices.append(pred_index[0])
writer.writerow([
method_name, k_fold, fold_counter,
len(pred_index[0][:k_max]),
len(true_index[0]), elapsed_time
] + pred_index[0][:k_max] + true_index[0])
# print("eval on test data.")
# prediction_test = autoencoder.predict(
# np.asarray([x_test_record[0].todense() for x_test_record in x_test]).reshape(x_test.__len__(), -1))
# y_test = np.asarray([y_test_record[0].todense() for y_test_record in y_test]).reshape(y_test.__len__(), -1)
# pred_indices, true_indices = dblp_eval.find_indices(prediction_test, y_test)
# for k in evaluation_k_set:
# # r@k evaluation
