0.2.3 (2021-06-09)

🚀 Features

• Add the skip_verify attribute to the KG class to skip or not the verification of the entity existence with remote Knowledge Graphs (default to skip_verify=False).
• Add WideSampler as a new sampling strategy.
• Add SplitWalker as a new walking strategy.

Fixed

• Fix the installation dependencies with poetry.
• Fix the cache memory for local Knowledge Graphs.
• Fix validation URL for remote Knowledge Graphs.
• Fix the HALKWalker walking strategy.
• Fix the DFS algorithm of RandomWalker and CommunityWalker to return duplicate walks and prevent a different number of walks for the entities.
• Fix the walk extraction with the with_reverse parameter for the different walking strategies.

Added

• Add the _post_extract private method in the Walker class for a post processing of walks by a walking strategy.

Changed

• Replace the default minimum frequency thresholds of a hop to keep with HALKWalker (0.001 -> 0.01).
• Drop support for Python 3.7.0
• Remove negative=20 and vector_size=500 for Word2Vec.

0.2.2 (2021-04-02)

🚀 Features

• Add a first support of FastText as embedding technique.

Fixed

• Fix the size hyperparameter by vector_size of the default dictionary in the Word2Vec class.
• Fix random determinism with walking strategies.
• Fix the calculation of walks for duplicate entities in a file.
• Fix the total recovery of entities, walks, literals and embeddings of a model after multiple online learning.

Added

• Add the _update private method in the RDF2VecTransformer class.
• Add the md5_bytes attribute in the CommuniWalker, HALKWalker, RandomWalker, and WLWalker classes to hash or not an object in MD5 and with how many bytes to keep.

Changed

• Replace the extract method in the Walker to returns a list of entities with their walks instead of a list of walks.

Fix the issue with nest-asyncio as dependency.

0.2.0 (2021-03-20)

🚀 Features

• Add support for Python 3.9
• Add the cache (default to cachetools.TTLCache(maxsize=1024, ttl=1200)) attribute to the KG class to significantly speed up the walks extraction through caching.
• Add the is_update (default to False) hyper-parameter in the fit method of the Embedder and Word2Vec classes to update an existing vocabulary.
• Add the literals (default to []) attribute in the KG class to support a basic literal extraction.
• Add the mul_req (default to False) attribute to the KG class to speed up the extraction of walks and literals for remote Knowledge Graph by sending asynchronous requests.
• Add the n_jobs (default to None) attribute to the Walker class to speed up the extraction of walks with multiprocessing.
• Add the random_state (default to None) parameter for the Walker class to handle better random determinism with walking and sampling strategies.
• Add the verbose (default to 0) attribute to the RDF2VecTransformer class to display useful debugging information and to measure the time of extraction, fit and generation of embeddings and literals.
• Add the with_reverse (default to False) parameter for the Walker class to generate more walks and improve the accuracy with Word2Vec, by including the parents of the entities in the walks.
• Add the possibility to do online learning of a model with the load and the save methods in the RDF2VecTransformer class.
• Add the validators for class parameter attributes.

Added

• Add the Connector generic class to simplify the implementation of new connectors.
• Add the SPARQLConnector class to delegate the connection part to the SPARQL endpoint server.
• Add the Vertex class in a slot to reduce RAM usage.
• Add the WalkerNotSupported and SamplerNotSupported exceptions in the Walker and Sampler classes when a walking strategy and a sampling strategy is not supported.
• Add the _cast_literals private method to the KG class to convert the raw literals of an entity according to their real types.
• Add the _embeddings, _entities, _literals, and _walks, attributes in the RDF2VecTransformer class to be able to get all the embeddings, entities, literals, and walks after the online training of a model.
• Add the _fill_hops private method in the KG class to fill the entity hops in cache when mul_req=True is provided for a remote Knowledge Graph.
• Add the _get_hops private method in the KG class to get the hops of a vertex for a local Knowledge Graph.
• Add the _is_support_remote (default to False) private attribute in the Walker and Sampler classes to restrict the use of walking and sampling strategies for some remote/local Knowledge Graph.
• Add the _res2hops private method in the KG class to convert a JSON response from a SPARQL endpoint server to hops.
• Add the add_walk method to the KG class to simplify the addition of walk in a Knowledge Graph.
• Add the attr decorator for all classes.
• Add the examples/online-training and examples/literals files to illustrate the use of online training and literals with pyRDF2Vec.
• Add the fetch_hops method to the KG class to fetch to get the hops of a vertex on a remote Knowledge Graph.
• Add the get_pliterals method to the KG class to gets the literals for an entity and a local KG based on a chain of predicates.
• Add the get_walks method in the RDF2VecTransformer class to get the walks of a given entities in a Knowledge Graph.
• Add the get_weights method in the Sampler class to get the hops weights.
• Add the pyrdf2vec.typings file to contains the aliases of the most commonly used typing with mypy.

Fixed

• Fix the get_weight method in the PageRankSampler to raise an error if the method is called before the fit method.
• Fix the remove_edge method of the KG class to also remove the edge of a children for a parent node.
• Fix the addition of predicate in memory for remote Knowledge Graphs.
• Fix the initialization of the _counts dictionary with the PredFreqSampler and ObjPredFreqSampler classes.

Changed

• Remove support for Python 3.6
• Remove the _get_shops and _get_rhops functions in the KG class.
• Remove the id attribute of the Vertex class.
• Remove the print_walks method of the Walker class.
• Remove the read_file method in the KG class.
• Remove the visualise method in the KG class.
• Replace the HalkWalker class by HALKWalker.
• Replace the SPARQLWrapper library in favor of using requests for synchronous requests and aiohttp for asynchronous requests.
• Replace the WeisfeilerLehmanWalker class by WLWalker.
• Replaces the add_edge, add_vertex, and remove_edge methods in the KG class to return a boolean value indicating that the addition/removal of an edge/vertex has been performed.
• Replace the depth parameter with max_depth for the Walker class.
• Replace the extract_random_community_walks, extract_random_community_walks_bfs, and extract_random_community_walks_dfs methods in the CommunityWalker class by extract_walks, _bfs, and _dfs methods.
• Replace the extract_random_walks, extract_random_walks_bfs, and extract_random_walks_dfs methods in the RandomWalker class by extract_walks, _bfs, and _dfs methods.
• Replace the file_type attribute in the KG class by fmt.
• Replace the get_inv_neighbors method in the KG class by a is_reverse (default to False) parameter in the get_neighbors method.
• Replace the initialize method in the Sampler class by the use of @property.
• Replace the is_remote parameter in the KG class for automatic link detection based on the http and https prefix.
• Replace the last parameter with is_last_depth in the sample_neighbor method of the Sampler class.
• Replace the label_predicates attribute in the KG class by skip_predicates and now use a set instead of a list.
• Replace the pyrdf2vec.graphs.kg.Vertex class with pyrdf2vec.graphs.Vertex.
• Replace the fit_transform and transform functions in the RDF2VecTransformer class to return a tuple containing the list of embeddings and literals.
• Replace the default embedding technique in the RDF2VecTransformer class for Word2Vec.
• Replace the default hyper-parameters of the Word2Vec class to size=500, min_count=0, and negative=20.
• Replace the default list of walkers in the RDF2VecTransformer class to [RandomWalker(2)].

Removes default prints from rdf2vec. Fix the README in PyPI.

Features

• Add a verbose (default to False) hyper-parameter for the fit method.
• Add basic support for remote Knowledge Graphs through SPARQL endpoint.
• Add configuration for Embedding Techniques through the Embedder abstract class (currently only Word2Vec is included).
• Add online documentation.
• Add sampling strategies (default to UniformSampler) from Cochez et al. to better deal with larger Knowledge Graphs.
• Add static typing for methods.
• Add support for Python 3.6 and 3.7.
• Add the Google Style Python Docstrings.
• Add the extract_random_walks_dfs and extract_random_walks_bfs methods for the RamdomWalker class.
• Add the get_hops method along with the private _get_rhops and _get_shops methods in the KG class.
• Add three examples (examples/countries.py, examples/mutag.py and examples/samplers.py) for pyRDF2vec.

Changed

• Replace graph for kg in the fit and fit_transform methods of the RDF2VecTransformer class.
• Replace instance for entities in the transform and fit_transform methods of the RDF2VecTransformer class.
• Replace default values of hyper-parameters of Word2Vec to match with the default ones of the gensim implementation.
• Replace the KnowledgeGraph class for KG.
• Replace the Walker class to be abstract.
• Replace the _rdf2vec.py file for rdf2vec.py.
• Replace the extract_random_community_walks method in the CommunityWalker to be private.
• Replace the extract methods in walkers to be private.
• Replace the graph.py file for graphs/kg.py.
• Replace the rdf2vec module for pyrdf2vec.
• Replace the imec licence for an MIT licence.
• Remove graph hyper-parameter in the transform method of the RDF2VecTransformer class.
• Remove hyper-parameters of RDF2VecTransformer for embedder and walkers ones.
• Remove the WildcardWalker walking strategy.
• Remove the converter.py file.
• Remove the create_kg, endpoint_to_kg, rdflib_to_kg functions for the location, file_type, is_remote hyper-parameters in KG with the read_file private method.
• Replace Vertex.vertex_count for itertools.count in the Vertex class.