Build feature analyzer in Python

[PGijsbers]

Create a Python-based feature analyzer that calculates basic feature information available through /data/features/{id} endpoint, e.g.:

{
  "data_features": {
    "feature": [
      {
        "index": "0",
        "name": "sepallength",
        "data_type": "numeric",
        "is_target": "false",
        "is_ignore": "false",
        "is_row_identifier": "false",
        "number_of_missing_values": "0"
      },
      {
        "index": "1",
        "name": "sepalwidth",
        "data_type": "numeric",
        "is_target": "false",
        "is_ignore": "false",
        "is_row_identifier": "false",
        "number_of_missing_values": "0"
      },
      {
        "index": "2",
        "name": "petallength",
        "data_type": "numeric",
        "is_target": "false",
        "is_ignore": "false",
        "is_row_identifier": "false",
        "number_of_missing_values": "0"
      },
      {
        "index": "3",
        "name": "petalwidth",
        "data_type": "numeric",
        "is_target": "false",
        "is_ignore": "false",
        "is_row_identifier": "false",
        "number_of_missing_values": "0"
      },
      {
        "index": "4",
        "name": "class",
        "data_type": "nominal",
        "nominal_value": [
          "Iris-setosa",
          "Iris-versicolor",
          "Iris-virginica"
        ],
        "is_target": "true",
        "is_ignore": "false",
        "is_row_identifier": "false",
        "number_of_missing_values": "0"
      }
    ]
  }
}

[Nidhicodes]

Hi @PGijsbers
Thanks for opening this, this looks like a very well-scoped and useful feature.

I’m interested in working toward a PR implementing the Python-based feature analyzer described here, as part of getting familiar with the OpenML server-api codebase and contributing more substantially going forward. Before starting implementation, I wanted to clarify expectations and constraints to make sure the approach aligns with maintainers’ vision.

My current understanding is that the goal is to:
-compute basic feature metadata (type, target/ignore flags, missing values, nominal values, etc.) from the dataset,
-and expose it in a form consistent with the /data/features/{id} endpoint contract.

A tentative approach I had in mind:
-implement a Python-side analyzer module that operates on the dataset representation already used server-side,
-clearly separate concerns between data loading, feature inspection, and response serialization,
-and ensure behavior matches the existing API output exactly (including edge cases like row identifiers, ignored features, and targets).

Happy to start with a small, focused PR and iterate based on feedback. My main goal is to understand the system well and contribute in a way that’s consistent with the project’s longer-term roadmap.

Looking forward to your thoughts!