Configuration¶

This page describes how to configure the data used to assess the model.

Options¶

Name	Allowed	Default	Description
`name`	`string`	`"isic2018"`	Name shown in outputs and tracking metadata.
`description`	`string, null`	`null`	Optional human-readable dataset description.
`source`	`string, null`	`null`	Path to a local data directory, or a named sample set such as `"imagenet_samples"`.
`labels.source`	`string, null`	`null`	Optional path to a labels file. Supported formats are CSV, TSV, and Parquet.
`labels.id_column`	`string, null`	`null`	Optional sample-ID column used to align labels with filenames, for example `"image"`.
`labels.column`	`string, null`	`null`	Optional class-label column. If omitted, one-hot numeric columns are reduced with `argmax`.
`labels.encoding`	`"index", "one_hot", "argmax", null`	`null`	Optional label parsing strategy.

YAML example¶

data:
  name: "my-dataset"
  description: "Internal validation set"
  source: "./data/images"
  labels:
    source: "./data/labels.csv"
    id_column: "image"
    column: "label"
    encoding: "index"

CLI override example¶

uv run raitap data.source="./data/images" data.labels.source="./data/labels.csv" data.labels.column=label

pip

raitap data.source="./data/images" data.labels.source="./data/labels.csv" data.labels.column=label