Lookup entity word embeddings given a flatten 1-D entity id list and content lookup table :param entities: Must be a 1-D entity vector :para

Content embedding without padding embeddings Works with 3-D tensor [?, content_len, embedding_size] and 4-D tensor [?, ?, content_len, emb

Based on training file, generate a subset of targets for each relationship. The assumption is if entity has relationship R before

:param transformed_heads: [?, ?, word_dim] :param transformed_rels: [?, word_dim] :param reuse: :param device:

:param content: [batch_size, ?, word_embedding_size] :param content_len: [batch_size] :return:

Change this to return a SparseTensor :param entity: entity must be a 1-D vector with one element? :param relation: :param targets_lookup

Generate head entity representation using given entity ids. :param heads: :param reuse: :param device: :param name:

Lookup a single entity's content from `content`, and convert words into ids using vocab_table. :param content: A 1-D string matrix :param vo

A convenience function for looking up both content and title embeddings. This will preserve the input entity shape. So the output shape would be

:param transformed_heads: [?, ?, word_dim] :param transformed_rels: [?, word_dim] :param reuse: :param device:

:param combined_head_rel: [?, ?, word_dim] :param tails: [?, ?, word_dim] :param reuse: :param device: :para

:param combined_head_rel: [?, ?, word_dim] :param tails: [?, ?, word_dim] :param reuse: :param device: :para

This is used to extract entity description and titles. :param heads: [?, ?] <- due to evaluation, sometimes heads will be (1, ?) but

Generate head entity representation using given entity ids. :param heads: :param reuse: :param device: :param name:

This is used to extract entity description and titles. :param heads: [?, ?] <- due to evaluation, sometimes heads will be (1, ?) but

:param rels: Any shape :param reuse: :param device: :param name: :return:

:param rels: Any shape :param reuse: :param device: :param name: :return:

:param rels: Any shape :param reuse: :param device: :param name: :return:

:param rels: Any shape :param reuse: :param device: :param name: :return:

:param tails: Any shape :param reuse: :param device: :param name: :return:

:param content: :param vocab_table: :param ids: :param name: :return: 2-D [batch_size, max_length_in_batch] content id matrix,

This is the transformation function for both head and tail entities :param ents: :param transformed_rels: :param reuse:

This is the transformation function for both head and tail entities :param ents: :param transformed_rels: :param reuse:

Corrupt the entity by __sampling from [0, max_range] and not in the true target set. :param triple: :param all_triples: :param corrupt_t

Calculate the averaging embedding of given entities. :param content_embedding: 3 dimension tensor with shape [?, embedding_size] :pa

Calculate the averaging embedding of given entities. :param content_embedding: 3 dimension tensor with shape [?, embedding_size] :pa

Run a series of sanity check functions to validate the model. :param entity_dict: A dictionary of {entity_str_name : entity_numerical_id}

Run a series of sanity check functions to validate the model. :param entity_dict: A dictionary of {entity_str_name : entity_numerical_id}

:param tails: Any shape :param reuse: :param device: :param name: :return:

Generate a set of training triples using a correct one. Input triple is a string triple, and the output elements are all numerical ids :par

Load normal head, tail, rel files, divide them into dicts { relation : { head : [tails] } } We also skip

Calculate the similarity :param entity_embeddings: [?, n_entities, content_length, word_embed_size] :param relation_embeddings: [batch_size,

Create all embedding matrices in this function. :param device: The storage device of all the embeddings. If you are

Create all embedding matrices in this function. :param device: The storage device of all the embeddings. If you are

Non trainable variables/constants. The variables created in this function will not be stored in the checkpoints. If the input data c

Non trainable variables/constants. The variables created in this function will not be stored in the checkpoints. If the input data c

Create an input pipeline that provides pre-processed, sampled training data The pipeline will be stored on CPU because it can not be opt

Create an input pipeline that provides pre-processed, sampled training data The pipeline will be stored on CPU because it can not be opt

Initialize all variables of the TF model. This should be called after the model is created, and no previous checkpoints are found. T

Initialize all variables of the TF model. This should be called after the model is created, and no previous checkpoints are found. T

Randomly corrupt head or tail with prob `head_corrupt_prob` :param triple: :param all_triples: :param max_entity_id: :param head_cor

Corrupt the relationship by __sampling from [0, max_range] :param triple: :param all_triples: :param max_range: :param name: :re

Extract an embedding for each instance. This is similar to DKRL, the first k-1 layers are using maxpooling and the last layer uses mean pool

Extract an embedding for each instance. :param content_embedding: [batch_size, n_entities, content_len, word_embedding] :param conv_p

Manually evaluate one single partial triple with a given set of targets This function will reduce the computation by reusing the targets of

Manually evaluate one single partial triple with a given set of targets :param device: :return:

Manually evaluate one single partial triple with a given set of targets This function will reduce the computation by reusing the targets of

Manually evaluate one single partial triple with a given set of targets This function will reduce the computation by reusing the targets of

Manually evaluate one single partial triple with a given set of targets This function will reduce the computation by reusing the targets of

Manually evaluate one single partial triple with a given set of targets This function will reduce the computation by reusing the targets of

Manually evaluate one single partial triple with a given set of targets This function will reduce the computation by reusing the targets of