Skip to content

CLI Reference

Full command surface. For usage examples see CLI & API.

Entry point: cli = "cli.main_cli:app" (Typer). --help works at every level.

cli data

Command Signature
download --type {pbmc,maxfuse,all} (default all)

cli preprocess

Command Signature
data DATASET-ID [--out/-o PATH]
info DATASET-ID [--out/-o PATH]
labels DATASET-ID [--out/-o PATH]
distances DATASET-ID [--out/-o PATH] [--verbose/-v]

DATASET-ID{cite_seq_and_pbmc, tonsils}. --out defaults to DEFAULT_PREPROCESSED_DIR_PATH.

cli generate

Command Signature
count SRC [--reps/-r N] (default 1)
new CONFIG_FILE OUT [--reps/-r N] (default 1)
batch SRC_DIR OUT_DIR BATCH_SIZE

cli run

train

cli run train RUN_TYPE CONFIGURATION-PATH
    [--id STR]                    default ''
    [--out/-o PATH]               default DEFAULT_OUT_DIR_PATH
    [--preprocessed-dir/-p PATH]  default None
    [--evaluate/--eval/-e]        default False
    [--verbose/-v]                default False

RUN_TYPE{original, regular, distributed, federated}.

evaluation / mrre

cli run evaluation DATASET_ID OUT-DATA-PATH
    [--preprocessed-dir/-p PATH]
    [--config-path/-c PATH]
    [--algorithm-type/-at TYPE]   default regular
    [--verbose/-v]

cli run mrre DATASET_ID OUT-DATA-PATH
    [--preprocessed-dir/-p PATH]
    [--config-path/-c PATH]
    [--algorithm-type/-al TYPE]   default regular
    [--verbose/-v]

Note

mrre is separate because it is the most memory-intensive metric: it needs full distance-rank matrices, scaling \(O(n^2)\).

cli analysis

Command Signature
matching PATH [--summary/-s]
embeddings PATH [--summary/-s]

cli export

Command Signature
images SRC [--output-dir/-o PATH] [--prefix/-p STR]
script SRC [--output-file/-o PATH]