ckpt_freq = int(steps_per_epoch / train_config['checkpoints_per_epoch'])
cp_callback = CheckpointSaver(ckpt_folder,
ckpt_freq,
steps_per_epoch,
starting_epoch=train_config['starting_epoch'])
sts_eval_callback = STSEvalCallback(
sts_val, sts_loss_weight=sts_config['sts_loss_weight'])
model = SentenceClassifier(cls_config['seq_len'], tokenizer,
train_config['rnn_units'],
train_config['recurrent_layer'],
train_config['embed_dim'])
model.set_sts_data(sts_train, 1, steps_per_epoch,
sts_config['sts_loss_weight'], sts_config['sts_batch_size'])
original_sentence_precision = ClassWisePrecision(
cls_config['original_sentence_label'], name='precision_original_sentence')
changed_sentence_precision = ClassWisePrecision(
cls_config['changed_sentence_label'], name='precision_changed_sentence')
original_sentence_recall = ClassWiseRecall(
cls_config['original_sentence_label'], name='recall_original_sentence')
changed_sentence_recall = ClassWiseRecall(cls_config['changed_sentence_label'],
name='recall_changed_sentence')
model.compile(optimizer=tf.keras.optimizers.Adam(
learning_rate=train_config['learning_rate']),
loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
metrics=[
