When convert a ml using backend of pytorch, how can I save the converted model into pytorch format?

[GoHard2023]

As titled. Is there a way I can save a converted model as pytorch format?

[interesaaat]

You can find the tests here showing how to do it with the different backends

[GoHard2023]

You can find the tests here showing how to do it with the different backends

Thank you for replying! I walked over the test cases. I did not find the test code to save converted model as pytorch format. Maybe I was not clear enough. I want to the saved model can be loaded by some pytorch code.

Thanks.

[interesaaat]

When you save a model we create a zip file containing a bunch of metadata and the model in the pytorch format (if you chose that as backend).

If you want just to save and load the pytorch model you can just do hb_model.model.save(path) and torch.load(path). (the model attribute is a pytorch model). Doing this way however you will lose the Hummingbird metadata so you will have to convert the input data yourself, for example.

[GoHard2023]

When you save a model we create a zip file containing a bunch of metadata and the model in the pytorch format (if you chose that as backend).

If you want just to save and load the pytorch model you can just do hb_model.model.save(path) and torch.load(path). (the model attribute is a pytorch model). Doing this way however you will lose the Hummingbird metadata so you will have to convert the input data yourself, for example.

Thanks for the tip! hb_model.model.save(path) does not work out for me. I used torch.save(hb_model.model.state_dict()) to save the model as pytorch match I guess,

[GoHard2023]

By the way, is there a way that I can select only probability when converting random forest to pytorch in hummingbird?

[interesaaat]

No. You can run ‘predict_prob’ to get the probabilities, but conversion is the same.

…

On Tue, Oct 18, 2022 at 8:16 PM GoHard2023 ***@***.***> wrote: By the way, is there a way that I can select only probability when converting random forest to pytorch in hummingbird? — Reply to this email directly, view it on GitHub <#641 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AALWR55TWGR5JCLQJNNX63DWD5RXHANCNFSM6AAAAAARHTWVKU> . You are receiving this because you commented.Message ID: ***@***.***>