Define Model signature for ML Flow

[stevehadd]

For registering our models in MLFlow (and probably other frameworks) we should define the schema for the model(s). This involves describing the following aspects:

• shape and data types of inputs
• tensor of columns based inputs

Defiining
https://www.mlflow.org/docs/latest/models.html#model-signature

The key code is as follows:

signature = mlflow.models.signature.infer_signature(testX, model.predict(testX))
mlflow.keras.log_model(model, "mnist_cnn", signature=signature)

It is not clear how this works with the autolog functionality.

[stevehadd]

this would be useful for a robist inference pipeline so we can save the trained models and easily use the mlflow for loading in the model and matching the data to essure we are doing inference correcftly for our testbed dashboard