def _get_preds_int(self, preds_vec):
"""Get integer predictions."""
utils.thin_line()
print('Converting prediction vectors to ints...')
preds = np.argmax(np.array(preds_vec), axis=1)
# Save preds
if self.cfg.SAVE_TEST_PRED:
if self.during_training and (self.epoch_train != 'end'):
utils.save_test_pred_is_training(self.test_log_path,
self.epoch_train,
self.step_train,
self.y_test,
preds,
preds_vec,
save_num=20,
pred_is_int=True)
else:
utils.save_test_pred(self.test_log_path,
self.y_test,
preds,
preds_vec,
pred_is_int=True)
return preds
def _get_preds_binary(self, preds_vec):
"""Get binary predictions.
-> [0, 0, 1, ..., 0, 1, 0] as labels
"""
utils.thin_line()
print('Converting prediction vectors to binaries...')
preds = np.array(preds_vec)
if self.cfg.MOD_PRED_MODE == 'top_n':
for pred_i in preds:
pos_idx = np.argsort(pred_i)[-self.cfg.MOD_PRED_MAX_NUM:]
neg_idx = np.argsort(pred_i)[:-self.cfg.MOD_PRED_MAX_NUM]
pred_i[pos_idx] = 1
pred_i[neg_idx] = 0
elif self.cfg.MOD_PRED_MODE == 'length_rate':
for pred_i in preds:
pred_i_copy = pred_i.copy()
max_ = pred_i.max()
pred_i[pred_i < (max_ * self.cfg.MOD_PRED_THRESHOLD)] = 0
pred_i[pred_i >= (max_ * self.cfg.MOD_PRED_THRESHOLD)] = 1
if np.sum(pred_i) > self.cfg.MOD_PRED_MAX_NUM:
pos_idx = np.argsort(
pred_i_copy)[-self.cfg.MOD_PRED_MAX_NUM:]
neg_idx = np.argsort(
pred_i_copy)[:-self.cfg.MOD_PRED_MAX_NUM]
pred_i[pos_idx] = 1
pred_i[neg_idx] = 0
else:
raise ValueError('Wrong Mode Name! Find {}!'.format(
self.cfg.MOD_PRED_MODE))
if self.cfg.SAVE_TEST_PRED:
if self.during_training and (self.epoch_train != 'end'):
utils.save_test_pred_is_training(self.test_log_path,
self.epoch_train,
self.step_train,
self.y_test,
preds,
preds_vec,
save_num=20)
else:
utils.save_test_pred(self.test_log_path, self.y_test, preds,
preds_vec)
return np.array(preds, dtype=int)