Separate method for Feature Engineering about featurewiz

[AutoViML]

Let me see what I can do - do you have any idea of what the syntax should look like for columns that need feature engg?

[inverse-panda]

some tools implement the feature_engg as simple methods with this signature,

def feature(x: List[Any], **params) -> List[float | bool]:
    pass

look at tsFresh. using this you can serialize the feature engg, all you need for each feature are:

1. the name of the feature engg method
2. the columns names that are input to the the feature
3. the extra parameters of the method

It gives the flexibility to user to run the feature engg and selection, in the training pipeline, and use the result to only and only run the feature engg on for the selected columns in the test pipeline.

[AutoViML]

hmmm... that's very interesting. In light of featurewiz, can you run a quick scan of the feature engg methods I have and name some functions that I should start with?If you remember, I already have scikit-learn compatible transformers for time series features, categorical transformations, etc. Those transformers should left as is I assume since they can be chained together in a pipeline using the sklearn Pipeline object. I am not sure if there are others. Let's work together and see what comes out.Auto Vimal On Thursday, July 6, 2023 at 07:31:25 AM EDT, inverse-panda ***@***.***> wrote: some tools implement the feature_engg as simple methods with this signature, def feature(x: List[Any], **params) -> List[float | bool]: pass look at tsFresh. using this you can serialize the feature engg, all you need for each feature are: - the name of the feature engg method - the columns names that are input to the the feature - the extra parameters of the method It gives the flexibility to user to run the feature engg and selection, in the training pipeline, and use the result to only and only run the feature engg on for the selected columns in the test pipeline. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>