Typed inputs, straight from your data.

schema.py

import pandas as pd

from mlschema import infer_schema

df = pd.DataFrame(
    {
        "score": [98.5, 86.0, 91.0],
        "role": pd.Categorical(["engineer", "engineer", "scientist"]),
        "active": [True, False, True],
    }
)

fields = infer_schema(df)

fields

[
  { "label": "score", "required": true, "mappedTo": "score", "kind": "number", "step": 0.1 },
  {
    "label": "role",
    "required": true,
    "mappedTo": "role",
    "kind": "category",
    "options": ["engineer", "scientist"]
  },
  { "label": "active", "required": true, "mappedTo": "active", "kind": "boolean" }
]

Every field is checked against a strict Pydantic model before it is returned. Constraints that contradict each other, unknown attributes and unknown kinds fail during inference instead of reaching your interface.

Find your path

Infer a first schema Shape a DataFrame, call infer_schema, and read the result. Know the field kinds Seven built-in kinds, how each is detected, and what it accepts. Reach the model mappedTo targets, one-hot groups, and positional feature vectors. Refine the result Labels, bounds, defaults and units, validated with the rest. Add your own rules Builders for existing kinds and kind() for new ones. Render a form Pass the fields to MLForm and submit the values the model expects.

MLschema is open source under the MIT license and part of eshkra. Install it from PyPI, try it in the playground, or read the source on GitHub.