Metadata helpers

The InsideForest.metadata module bridges cluster descriptions with your organization’s data catalog so you can audit recommendations and design follow-up experiments.

Main utilities

• MetaExtractor: maps the tokens found in cluster_description to rows in a metadata catalog and filters the columns to keep via the Profile enum.
• parse_rule_string, token_from_condition and conditions_to_tokens: helpers that split rule text into individual conditions and reusable tokens.
• experiments_from_df2 and run_experiments: generate contrastive hypotheses between clusters, ranking where to intervene first.

Example

import pandas as pd
from InsideForest import MetaExtractor, Profile, run_experiments

metadata_df = profiler_table.set_index("variable_id")  # must include the target row
mx = MetaExtractor(metadata_df, var_obj="target")

catalog = mx.extract(
    clf.df_clusters_description_,
    profile=Profile.INVESTIGATION,
)

records = training_df.copy()
records["target"] = y
hypotheses = run_experiments(
    mx,
    {"iris": clf.df_clusters_description_},
    data_dict={"iris": records},
)

For examples with metadata.df.head(10), df_meta_sub, top_experiments, and generated hypotheses, review full pipeline outputs.

Note: the metadata catalog must be indexed by the variable tokens found in cluster_description plus the target column. The detailed DataFrames df_clusters_description_ and df_datos_explain_ are populated when the estimator is fitted with get_detail=True. Combine the extracted catalog with reproducibility assets from the Reproducibility guide.