Command Line Interface

Command line utility to run pdstools apps.

usage: pdstools [-h] [--version] [--list] [--data-path DATA_PATH]
                [--sample SAMPLE] [--filter FILTER] [--temp-dir TEMP_DIR]
                [--full-embed] [--no-full-embed]
                [{launcher,health_check,decision_analyzer,impact_analyzer,all,hc,da,ia}]

Positional Arguments

app

Possible choices: launcher, health_check, decision_analyzer, impact_analyzer, all, hc, da, ia

The app to run: “launcher” (alias: all) | “health_check” (alias: hc) | “decision_analyzer” (alias: da) | “impact_analyzer” (alias: ia)

Named Arguments

--version

show program’s version number and exit

--list

List available apps (one per line, tab-separated key/name/aliases) and exit.

Default: False

--data-path

Path to a data file or directory to load on startup. Per-app expectations: DA — single file or directory of explanations export (parquet, csv, json, arrow, zip, tar, partitioned folders); HC — directory or zip containing an ADM Datamart export (model snapshot + predictor binning, optional prediction table); IA — single JSON/NDJSON (PDC), ZIP (VBD), or XLSX file. Exposed to the app as the PDSTOOLS_DATA_PATH env var.

--sample

Pre-ingestion interaction sampling for large datasets (applies to DA and IA; no-op for HC). Specify an absolute count (e.g. ‘100000’, ‘100k’, ‘1M’) or a percentage (e.g. ‘10%’). All rows for each sampled interaction are kept. Exposed to the app as the PDSTOOLS_SAMPLE_LIMIT env var. To sample programmatically without the app, see pdstools.decision_analyzer.utils.sample_interactions() and pdstools.decision_analyzer.utils.prepare_and_save().

--filter

Pre-ingestion row filter for extracting specific data from large files (DA only; no-op for HC and IA). Syntax options: ‘Column=value1,value2,…’ (categorical, exact match), ‘Column>=N’ / ‘Column<=N’ / ‘Column>N’ / ‘Column<N’ (numeric), ‘Column=YYYY-MM-DD..YYYY-MM-DD’ (date range, inclusive). Column names use display names (e.g. ‘Channel’, ‘Decision Time’, ‘ModelPositives’). Multiple –filter flags are ANDed together. Can be combined with –sample (filter is applied first).

--temp-dir

Directory for temporary files such as the sampled data parquet (applies to DA and IA; no-op for HC). Defaults to the current working directory. Exposed to the app as the PDSTOOLS_TEMP_DIR env var.

--full-embed

Bundle all JS/CSS libraries (Plotly, itables, etc.) directly into the generated HTML report, producing a fully self-contained file that works offline and in air-gapped environments (HC only; no-op for DA and IA). The file will be larger and esbuild is required. Without this flag (default) libraries are loaded from CDN — smaller file, but requires an internet connection at viewing time. Exposed to the app as the PDSTOOLS_FULL_EMBED env var.

Default: False

--no-full-embed

Load JS/CSS libraries from CDN (default; HC only). Produces a smaller report file but requires internet access when viewing. Use –full-embed for offline/air-gapped environments.

Default: True