Skip to content

cli.md

Latest commit

 

History

History
71 lines (52 loc) · 2.27 KB

cli.md

File metadata and controls

71 lines (52 loc) · 2.27 KB
Raw

CLI reference

prefscope --help
prefscope COMMAND --help

The CLI prepares data, builds lenses, extracts features, and runs interpretation tools. Analysis and report orchestration are direct Python workflows and do not have registry- or config-driven CLI commands.

Data

  • prefscope inspect — inspect a corpus or annotation table.
  • prefscope init-demo — create a local demo dataset.
  • prefscope build-corpus — normalize a paired corpus.
  • prefscope prepare-dataset — map a table into PrefScope's canonical row form.

Native lens construction

  • prefscope build-lens
  • prefscope embed-corpus
  • prefscope embed-prompts
  • prefscope build-prompt-lens
  • prefscope package-lens
  • prefscope package-bundle — package several independent child lenses under one explicit bundle manifest.

Use each command's --help output for its alpha flags. Native lens training can require optional model and Torch dependencies.

Feature extraction and interpretation

  • prefscope encode-dataset
  • prefscope interpret name
  • prefscope interpret verify

These commands create or annotate lens artifacts. Proposed names do not by themselves establish feature meaning or fidelity.

Token SAE tools

  • prefscope extract-activations
  • prefscope train-token-sae
  • prefscope summarize-activations

These commands are optional training utilities and can require heavy dependencies.

Viewer bridge

The separate entry point packages an already saved FeatureBatch with an explicit built Viewer directory:

prefscope-export-viewer \
  --features artifacts/features \
  --viewer-dist path/to/viewer/dist \
  --catalog artifacts/feature_catalog.json \
  --table activity=results/activity.csv \
  --out results/viewer-site

--viewer-dist must contain a compatible root viewer-build.json. --out is a new self-contained static-site directory and must not exist. The command does not calculate an analysis, build the Viewer, or start a server.

Removed commands

Commands that dispatched fixed analyses, compared responses automatically, generated reports, classified roles, or ran YAML workflows were removed. Use the direct numerical functions in Python API, and call any specialized recipe explicitly when a study needs it.