Is sklearn pipeline via river submodules possible

[prabhant]

If i have something like a river scaler and classifier then is it possible to put them together in a sklearn pipeline? Or do i need to convert the entire river pipeline then only i can arrange it like that?
If its possible then can you give an example because i was not able to make it work by myself

[MaxHalford]

Have you checked out the compat module? It has some utilities for wrapping a River model to make it compatible with scikit-learn, and vice-versa.

[prabhant]

Is there also any way to change the hyperparameters of the converted model?

[MaxHalford]

You should be able to use model._set_params and provide a nested dict with the params you want. It depends on what you want to do exactly.

[medha-chippa]

@prabhant hello, did you use the river_to_sklearn() from the compact module to convert your river model to the one compatible with scikit learn.
I've been wanting to convert a model in river to the one that is compatible with scikitlearn. But I've not been able to do this. It doesn't seem to work. Can you please guide me in this regard.
Thank you

[prabhant]

@medha-chippa you can ping me on my email (you can find it in the github profile
)

[prabhant]

@MaxHalford I tried the model.set_params
This was my river to skl model
River2SKLClassifier(river_estimator=LogisticRegression ( optimizer=SGD ( lr=Constant ( learning_rate=0.01 ) ) loss=Log ( weight_pos=1. weight_neg=1. ) l2=0. intercept_init=0. intercept_lr=Constant ( learning_rate=0.01 ) clip_gradient=1e+12 initializer=Zeros () ))

Now i wanted to change the hyperparameters like loss and l2
but when i check the keys available with get_params i get
{'river_estimator': LogisticRegression ( optimizer=SGD ( lr=Constant ( learning_rate=0.01 ) ) loss=Log ( weight_pos=1. weight_neg=1. ) l2=0. intercept_init=0. intercept_lr=Constant ( learning_rate=0.01 ) clip_gradient=1e+12 initializer=Zeros () )}

and the only key here is river estimator. So I can't really change the l2 directly via model.set_params. Is this the intended behavior?