API reference¶

Everything below is importable directly from the top-level themisim package.

Querying¶

`query`	Find the frames most similar to `(site, datetime, frame)`.
`resolve_global_id`	Map `(site, datetime, frame)` to the indexed vector's global id.
`SearchEngine`	Loads the index + manifest + memmap once; serves many queries.
`Hit`

Exporting results¶

`hits_to_dataframe`	The same four columns the dashboard results grid shows, same order.
`results_csv_filename`	`themis-similarity-search-<site>-<YYYYMMDDTHHMMSSZ>.csv` for an in-index query frame; a plain fallback otherwise.

Building an index¶

`download_archive`	Crawl the archive for the requested sites/dates and download the CDFs.
`build_index`	Embed every downloaded CDF, concatenate, and build the FAISS index.
`fetch_weights`	Return a local path to the verified checkpoint, downloading if needed.

Pilot indexes¶

A pilot is a small, pinned slice of the real archive that builds in minutes and reports its own retrieval quality, so the software can be assessed without the full ~100 TB archive. See Pilot indexes.

`list_pilots`	Every registered pilot spec, smallest first.
`build_pilot`	Download, build and (by default) validate one pilot slice.
`validate_index`	Run every check against `artifacts` and return the report dict.
`run_benchmark`	Full report: environment, hyperparameters, latency, throughput, recall.

Visualization¶

visualize_results

Render query results in a grid, like the dashboard's results view.