Problem with Adam optimizer while using MLPRegressor.

[tamoghnabh]

I am trying to fit a regression model using MLPRegressor module. It works fine with SGD as the optimizer. But I get the following error 'numpy.ndarray' object has no attribute 'items' if I use Adam as the optimizer. Does this problem has something to do with the statement in the API reference stating "This model is still work in progress." Is there any workaround as I am expecting improvement in performance with Adam?

[MaxHalford]

Indeed not all of the optimizers can handle both numpy arrays and dicts. @gbolmier had started to adapt the optimizers in #338, but it never got finished.

[MaxHalford]

@Geronimo1995 we're adding support for Adam in #722

[tamoghnabh]

That is fantastic news. Expecting significant performance improvement in my model.

[MaxHalford]

It's merged. You can install the latest development version to use it.

[tamoghnabh]

It seems that the problem persists even after updating to the latest version.

[MaxHalford]

Ok I found the issue and I just fixed it. I did a dirty hack allowing each layer to have its own optimizer. TBD if it's the best way to go. Try again and it should work!

[tamoghnabh]

Worked like a CHARM!! Thank you so much.