minimel.experiment module

class minimel.experiment.log_time(fname)

Bases: object

minimel.experiment.find(directory, glob)

minimel.experiment.sweep(**kw)

minimel.experiment.make_dir_params(name, **params)

minimel.experiment.get_dir_params(dirname: Path)

Parameters:

dirname (Path)

minimel.experiment.experiment(root: Path = PosixPath('.'), *, outdir: Path | None = None, nparts: int = 100, head: int | None = None, split: List[int] = (None,), fold: List[int] = (None,), stem: List[str] = ('',), min_count: List[int] = (2,), freqnorm: List[bool] = (False,), badentfile: List[Path] = ('',), tokenscore_threshold: List[float] = (0.1,), entropy_threshold: List[float] = (1.0,), countratio_threshold: List[float] = (0.5,), quantile_top_shadowed: List[float] = (0,), cluster_threshold: List[float] = (None,), vectorizer: List[Path] = ('',), ent_feats_csv: List[Path] = ('',), balanced: List[bool] = (False,), usenil: List[bool] = (False,), bits: List[int] = (20,), runfile: List[Path] = ('',), use_fallback: List[bool] = (True,), also_baseline: bool = True, evaluate: bool = False, evaluate_per_name: bool = False)

Run all steps to train and evaluate EL models over a parameter sweep.

The root directory must contain the following files:

• index_*.dawg: DAWG trie mapping of article names -> numeric IDs

• *-disambig.txt: See disambig_ent_file in get_disambig

Parameters:

• root (Path) – Root directory

• outdir (Optional[Path])

• nparts (int)

• head (Optional[int])

• split (List[int])

• fold (List[int])

• stem (List[str])

• min_count (List[int])

• freqnorm (List[bool])

• badentfile (List[Path])

• tokenscore_threshold (List[float])

• entropy_threshold (List[float])

• countratio_threshold (List[float])

• quantile_top_shadowed (List[float])

• cluster_threshold (List[float])

• vectorizer (List[Path])

• ent_feats_csv (List[Path])

• balanced (List[bool])

• usenil (List[bool])

• bits (List[int])

• runfile (List[Path])

• use_fallback (List[bool])

• also_baseline (bool)

• evaluate (bool)

• evaluate_per_name (bool)

Keyword Arguments:

• outdir – Write outputs to this directory

• nparts – Number of parts to chunk wikidump into

• head – Use only N first lines from each partition

• split – Split the data into several parts

• fold – Ignore this fold of the split data in training, use in evaluation

• stem – Stemming language ISO 639-1 (2-letter) code (use X for no stemming)

• min_count – Minimal (anchor-text, target) occurrence

• freqnorm – Normalize counts by total entity frequency (1/0)

• badentfile – File of entity IDs to ignore, one per line (default: *-disambig.txt)

• tokenscore_threshold – Threshold for mean asymmentric Jaccard index between name and candidate entity labels

• entropy_threshold – Entropy threshold (high entropy = flat dist)

• countratio_threshold – Count-ratio (len / sum) threshold

• quantile_top_shadowed – Only train models for a % names with highest counts of candidate entities shadowed by the top candidate

• cluster_threshold – Cluster names based on their meanings

• vectorizer – Scikit-learn vectorizer .pickle or Fasttext .bin word embeddings. If unset, use tokens directly.

• ent_feats_csv – CSV of (ent_id,space separated feat list) entity features

• balanced – Use balanced training

• usenil – Use NIL option for training unlinked mentions

• bits – Number of bits of the Vowpal Wabbit feature hash function

• runfile – TSV rows of (ID, {name -> ID}, text) or ({name -> ID}, text)

• use_fallback – Use raw counts as fallback

• also_baseline – Also run a baseline model without model predictions

• evaluate – Write evaluation scores to file

• evaluate_per_name – Write evaluation scores per name to file