Re: Infer_signature for a dictionary datasets duri...

Kumaran · ‎08-21-2025

Thank you for contacting Databricks support!

Because infer_signature() cannot handle a dict of DataFrames directly, you will need to convert this structure into a dictionary of row dicts, or manually build a ModelSignature.

Option A: Flatten Each DataFrame to Dict

# Create a dictionary of dicts (each containing a single record for each model)
input_example_simplified = {
    k: v.iloc[0].to_dict()
    for k, v in input_example.items()
}

Now input_example_simplified looks like:

{
    "local_outlier_factor": {'x': 1.23, 'y': 4.56},
    "isolation_forest": {'a': 0.1, 'b': 0.2, 'c': 0.3}
}

You can use this with infer_signature():

from mlflow.models.signature import infer_signature

signature = infer_signature(input_example_simplified, output_example_simplified)