graph-preloader

Transforming triples into objects and objects into better objects

running

This project uses the pipenv tool to manage the virtual environment. To run the scripts included here, install pipenv if needed, install dependencies, and then enter an environment shell:

pip install pipenv
pipenv install --dev
pipenv shell

single commands can also be run in the enviroment without spawning a full shell:

pipenv run my_script.py

check the pipenv documentation for instructions on how to add or remove requirements.

wikidata

There are several scripts for pulling, separating, and transforming wikidata objects in the wikidata directory. They all come with a short summary at the top. Scripts that require input take it from stdin. Output is written to stdout, unless output needs to be separated into multiple files, in which case a data/ directory is expected where output files can be written to. Specifying output filenames is a possible future enhancement.

dbpedia

The DBpedia preloader tool can be used as follows:

pipenv run python -m dbpedia.preloader -h
usage: preloader.py [-h] [--parallel] [--shorten-uris]
                    [--target-size TARGET_SIZE]
                    [--global-id-marker GLOBAL_ID_MARKER]
                    [--id-marker-prefix ID_MARKER_PREFIX]
                    [--parts-file PARTS_FILE] [--task-timeout TASK_TIMEOUT]
                    [--search-type {binary,jump}]
                    [--bin-search-limit BIN_SEARCH_LIMIT]
                    [--jump-size JUMP_SIZE] [--backpedal-size BACKPEDAL_SIZE]
                    [input_path] [output_dir]

Transform sorted Databus NTriples into property graph-friendly JSON.

positional arguments:
  input_path            the Databus NTriples input file path (default:
                        graph-preloader/dbpedia/sorted.nt)
  output_dir            the JSON output directory path (default:
                        graph-preloader/dbpedia/output_{hex}/)

optional arguments:
  -h, --help            show this help message and exit
  --parallel            transform parts in parallel using a multiprocessing
                        pool (default: False)
  --shorten-uris        shorten URIs by replacing known namespaces with their
                        corresponding prefix (default: False)
  --target-size TARGET_SIZE
                        the approximate size of parts in bytes (default: 500e6)
  --global-id-marker GLOBAL_ID_MARKER
                        only triples with this marker in the subject will be
                        transformed (default: id.dbpedia.org/global/)
  --id-marker-prefix ID_MARKER_PREFIX
                        the characters that precede the `global_id_marker` in
                        each triple (default: <http://)
  --parts-file PARTS_FILE
                        the file in which output files are listed with
                        corresponding input file positions (left and right)
                        (default: <output_dir>/parts.tsv)
  --task-timeout TASK_TIMEOUT
                        the number of seconds a "transform part" task is
                        allowed to run (applies only to parallel execution)
                        (default: 600)
  --search-type {binary,jump}
                        the type of search to use to skip to the first
                        `global_id_marker` triple (default: binary)
  --bin-search-limit BIN_SEARCH_LIMIT
                        the maximum number of iterations of the binary search
                        main loop (default: 120)
  --jump-size JUMP_SIZE
                        the size of forward jumps in bytes (default: 350e6)
  --backpedal-size BACKPEDAL_SIZE
                        the size of backpedals in bytes (default: <jump_size> // 10)