Can I pass a feature vector as a param ?

[msayan-vcbs]

Hi, so I have a desire to use this plugin in an unconventional way. I want to use a pre-trained xgboost model and provide a one hot encoded DenseFeatureVector as a query param which will be used in rescoring all my results, rather than use the FeatureSet framework paradigm.

Is this possible ?

Basically something like:

"rescore": {
    "window_size": 1000,
    "query": {
        "rescore_query": {
            "sltr": {
                "params": {
                    "keywords": "rambo"
                },
                "model": "my_model",
                "active_features": [0.1, 0.2, 0.4, 0.2, ... etc]
            }
        }
    }
}

}

[nathancday]

you should be able to pass in various JSON-y objects through params:, don't think functions would work but arrays should

[msayan-vcbs]

Hi Nathan, thanks for the quick reply.

So looking at the documentation, it looks like it may be possible to 'override' features through the Script Features Param section, correct ?

Would I have to define dummy fields in the feature set to achieve something like what I posted ?

[nathancday]

I believe a model requires a featureset to be created, so you'd need at least a dummy featureset like you said. Since a model's features already accept params, you'd just be calling sltr with active_features under the params key at the same level at keywords