MLFlow Output Signature

[pablobd]

I want to define the output of a MLFlow model to include a dictionary of string to float. I did this, but it doesn't work. Do you know if it's possible at all?

from mlflow.types.schema import Schema, ColSpec

output_schema = Schema([ ColSpec("dictionary_output", "map[string,float]") ])

[Kaniz · Accepted solution]

Hi @pablobd, The issue you’re facing might be due to the fact that MLflow’s ColSpec does not directly support a dictionary or map data type. The ColSpec class is used to specify the type and optionally the name of a single column in the data....

However, you can potentially work around this by having your model output a string representation of the dictionary, and then convert this string back to a dictionary in your application code. 

In this example, the model’s output is a JSON string, which is a string representation of the dictionary. This string can be easily converted back to a dictionary using json.loads() in your application code.

If you have further questions or need more help, feel free to ask!

[pablobd]

Solved with the json module that converts the dict into string, thanks! 🙂

import json

json.dumps(...)