def minibatch_to_feed_dict(self, batch_data: Dict[str, Any],
feed_dict: Dict[tf.Tensor,
Any], is_train: bool) -> None:
super().minibatch_to_feed_dict(batch_data, feed_dict, is_train)
write_to_feed_dict(feed_dict, self.placeholders['tokens'],
batch_data['tokens'])
write_to_feed_dict(feed_dict, self.placeholders['tokens_mask'],
batch_data['tokens_mask'])
def minibatch_to_feed_dict(self, batch_data: Dict[str, Any], feed_dict: Dict[tf.Tensor, Any], is_train: bool) -> None:
super().minibatch_to_feed_dict(batch_data, feed_dict, is_train)
feed_dict[self.placeholders['rnn_dropout_keep_rate']] = \
self.get_hyper('rnn_dropout_keep_rate') if is_train else 1.0
feed_dict[self.placeholders['rnn_recurrent_dropout_keep_rate']] = \
self.get_hyper('rnn_recurrent_dropout_keep_rate') if is_train else 1.0
write_to_feed_dict(feed_dict, self.placeholders['tokens'], batch_data['tokens'])
write_to_feed_dict(feed_dict, self.placeholders['tokens_lengths'], batch_data['tokens_lengths'])
def minibatch_to_feed_dict(self, batch_data: Dict[str, Any], feed_dict: Dict[tf.Tensor, Any],
is_train: bool) -> None:
super().minibatch_to_feed_dict(batch_data, feed_dict, is_train)
node_masks = batch_data['node_masks']
node_token_ids = batch_data['tokens']
if node_masks:
max_tokens = max([len(x) for x in node_masks])
node_masks = [n + [0] * (max_tokens - len(n)) for n in node_masks]
node_token_ids = [n + [-1] * (max_tokens - len(n)) for n in node_token_ids]
tfutils.write_to_feed_dict(feed_dict, self.placeholders['node_masks'], node_masks)
tfutils.write_to_feed_dict(feed_dict, self.placeholders['tokens'], node_token_ids)
def minibatch_to_feed_dict(self, batch_data: Dict[str, Any],
feed_dict: Dict[tf.Tensor,
Any], is_train: bool) -> None:
super().minibatch_to_feed_dict(batch_data, feed_dict, is_train)
node_type_ids = batch_data['node_type_ids']
children = batch_data['children']
node_type_ids, children = tbcnn_network.pad_batch(
node_type_ids, children)
tfutils.write_to_feed_dict(feed_dict,
self.placeholders['node_type_ids'],
node_type_ids)
tfutils.write_to_feed_dict(feed_dict, self.placeholders['children'],
children)
def minibatch_to_feed_dict(self, batch_data: Dict[str, Any], feed_dict: Dict[tf.Tensor, Any], is_train: bool) -> None:
super().minibatch_to_feed_dict(batch_data, feed_dict, is_train)
if self.get_hyper('use_token_embeddings'):
write_to_feed_dict(feed_dict, self.placeholders['token_embeddings'], batch_data['token_embeddings'])
else:
write_to_feed_dict(feed_dict, self.placeholders['tokens'], batch_data['tokens'])
write_to_feed_dict(feed_dict, self.placeholders['tokens_mask'], batch_data['tokens_mask'])
def minibatch_to_feed_dict(self, batch_data: Dict[str, Any], feed_dict: Dict[tf.Tensor, Any],
is_train: bool) -> None:
super().minibatch_to_feed_dict(batch_data, feed_dict, is_train)
node_masks = batch_data['node_masks']
node_token_ids = batch_data['node_token_ids']
edges = batch_data['edges']
# seq_masks = batch_data['seq_masks']
# seq_token_ids = batch_data['seq_token_ids']
if node_masks:
# pad batches so that every batch has the same number of nodes
max_tokens = max([len(x) for x in node_masks])
node_masks = [list(n) + [0] * (max_tokens - len(n)) for n in node_masks]
node_token_ids = [list(n) + [-1] * (max_tokens - len(n)) for n in node_token_ids]
edges = [
[edge_type, batch_id, v, u]
for batch_id, edge_pack in enumerate(edges)
for edge_type, v, u in edge_pack
if v < max_tokens and u < max_tokens
]
tfutils.write_to_feed_dict(feed_dict, self.placeholders['node_masks'], node_masks)
tfutils.write_to_feed_dict(feed_dict, self.placeholders['node_token_ids'], node_token_ids)
tfutils.write_to_feed_dict(feed_dict, self.placeholders['edges'], edges)
