loaders/

Reading and writing datasets, in whatever format they arrive.

Every loader returns the same Dataset — X, y and feature names — so the rest of TuiML never has to care where the data came from. Each format also has a matching save_*.

Formats

CSV / TSV: load_csv, save_csv.
ARFF: load_arff, save_arff. A typed text format that

carries column types and nominal values in its header.

Parquet: load_parquet, save_parquet, plus

load_parquet_partitioned for directory-partitioned datasets.

Excel: load_excel, save_excel, and

load_excel_sheets for a workbook of several sheets.

JSON: load_json, load_jsonl, load_json_nested

for records that are not flat, and their save_* counterparts.

NumPy: load_numpy, save_numpy (.npy / .npz).
pandas: from_pandas, to_pandas for in-memory frames.

Detecting the format

load and save pick the right one from the file extension, so a path is usually all you need. This is what lets {"source": "sales.csv"} work in a train spec.

python

>>> from tuiml.datasets.loaders import load
>>> data = load("sales.csv", target="label")     # doctest: +SKIP
>>> data.X.shape                                 # doctest: +SKIP
(1000, 12)