def get_regression_loss(args, xlnet_config, features, is_training=False):
"""Loss for downstream regression tasks."""
inp = fluid.layers.transpose(features["input_ids"], [1, 0, 2])
seg_id = features["segment_ids"]
inp_mask = fluid.layers.transpose(features["input_mask"], [1, 0])
label = features["label_ids"]
xlnet_model = XLNetModel(input_ids=inp,
seg_ids=seg_id,
input_mask=inp_mask,
xlnet_config=xlnet_config,
args=args)
summary = xlnet_model.get_pooled_out(args.summary_type, args.use_summ_proj)
per_example_loss, logits = modeling.regression_loss(
hidden=summary,
labels=label,
initializer=_get_initializer(args),
name="model_regression_{}".format(args.task_name.lower()),
return_logits=True)
total_loss = fluid.layers.reduce_mean(per_example_loss)
return total_loss, per_example_loss, logits
def get_regression_loss(FLAGS, features, is_training):
"""Loss for downstream regression tasks."""
bsz_per_core = tf.shape(features["input_ids"])[0]
inp = tf.transpose(features["input_ids"], [1, 0])
seg_id = tf.transpose(features["segment_ids"], [1, 0])
inp_mask = tf.transpose(features["input_mask"], [1, 0])
label = tf.reshape(features["label_ids"], [bsz_per_core])
xlnet_config = xlnet.XLNetConfig(json_path=FLAGS.model_config_path)
run_config = xlnet.create_run_config(is_training, True, FLAGS)
xlnet_model = xlnet.XLNetModel(xlnet_config=xlnet_config,
run_config=run_config,
input_ids=inp,
seg_ids=seg_id,
input_mask=inp_mask)
summary = xlnet_model.get_pooled_out(FLAGS.summary_type,
FLAGS.use_summ_proj)
with tf.variable_scope("model", reuse=tf.AUTO_REUSE):
per_example_loss, logits = modeling.regression_loss(
hidden=summary,
labels=label,
initializer=xlnet_model.get_initializer(),
scope="regression_{}".format(FLAGS.task_name.lower()),
return_logits=True)
total_loss = tf.reduce_mean(per_example_loss)
return total_loss, per_example_loss, logits
def get_regression_loss(FLAGS, features, n_class, is_training, mode):
bsz_per_core = tf.shape(features["input_ids"])[0]
inp = tf.transpose(features["input_ids"], [1, 0])
seg_id = tf.transpose(features["segment_ids"], [1, 0])
inp_mask = tf.transpose(features["input_mask"], [1, 0])
label = tf.reshape(features["label_ids"], [FLAGS.train_batch_size, 31])
xlnet_config = xlnet.XLNetConfig(json_path=FLAGS.model_config_path)
run_config = xlnet.create_run_config(is_training, True, FLAGS)
xlnet_model = xlnet.XLNetModel(xlnet_config=xlnet_config,
run_config=run_config,
input_ids=inp,
seg_ids=seg_id,
input_mask=inp_mask)
summary = xlnet_model.get_pooled_out(FLAGS.summary_type,
FLAGS.use_summ_proj)
# if mode == tf.estimator.ModeKeys.TRAIN or mode == tf.estimator.ModeKeys.EVAL:
with tf.variable_scope("model", reuse=tf.AUTO_REUSE):
per_example_loss, logits = modeling.regression_loss(
mode=mode,
hidden=summary,
labels=label,
n_class=n_class,
initializer=xlnet_model.get_initializer(),
scope="regression_{}".format(FLAGS.task_name.lower()),
return_logits=True)
total_loss = tf.reduce_mean(per_example_loss)
return total_loss, per_example_loss, logits
