Kglab

2021-04-23

* add conda `environment.yml` plus instructions
* use `bandit` to check for security issues
* add `pre-commit` hooks
* add `lgtm` code reviews (slow, not integrating into GitHub PRs directly)
* add use of `pipdeptree` to analyze dependencies
* refactor docs to use Divio design patterns; kudos louisguitton
* created `SECURITY.md` security policy
* using Snyk for security audits of dependencies (it has some troubles)
* testing/correcting the PSL examples; kudos louisguitton
* include first user quote; kudos kaaloo !!
* multifile decorator (supports globs) for `load_rdf()`; kudos Ankush-Chander
* fixed bug in SubgraphTensor null check; kudos dvsrepo
* prep for pytorch_geometric integration; kudos dvsrepo
* add `as_tensor()` support for PyTorch

2021-03-12

* tutorial proofreading and corrections; kudos louisguitton
* better exception handling for RDF construction, parsing, serialization
* `build_nx_graph()` now requires a `networkx.DiGraph` directed graph as an argument
* add DOI to README, CITATION, and docs
* add contributor instructions
* neo4j integration via `neosemantics`; kudos gauravjaglan
* replace `GPUtil` with `pynvml` as recommended by NVidia; kudos kingmesal
* option to enable NVidia GPU use, if devices are available and libraries are installed
* using `cuDF`/`cuGraph` in lieu of `pandas`/`networkx` where possible; kudos BradReesWork

2021-02-07

* project milestone release
* begin integration of `csvwlib`

2021-01-31

* hotfix: backing out the `igraph` dependency; too many installation issues with it
* factory pattern to build an `igraph` object allows caller to create and pass in, instead

2021-01-30

* fix for `save_parquet()` since `storage_options` won't be used by `pandas` before 1.2.0
* introduced `SubgraphMatrix` and `SubgraphTensor` as subclasses for the base vectorization
* introduced `PSLModel` class for building *probabilistic soft logic* models
* using `icecream` in methods that require verbose traces

2021-01-19

* rollback `pandas` to 1.1.5 which is the latest stable on Ubuntu; see 94
* revise `pyvis` use as a factory pattern
* extend RDFlib serialization with `urlpath` support
* extend all serialization to support file-like objects
* including `gcsfs` and `fsspec` to leverage cloud bucket access in `pandas`
* added `code_of_conduct.md`
* code checking based on `pylint`
* added accessor methods, e.g., `KnowledgeGraph.rdf_graph()`
* updated `pandas` usage for direct Parquet/Arrow calls, e.g., `read_parquet()` and `to_parquet()`
* updated `pyshacl` usage to >= 0.14.1 to support `inplace` graph manipulation (inference)