How save model in file?

[axsapronov]

Hi all,
I took example (from docs), and i want to save classifier to file.
But if:

with open('dump_autosk.pkl', 'wb') as fio:
    pickle.dump(automl, fio)

I got

TypeError: Pickling an AuthenticationString object is disallowed for security reasons

Can i save model in file?

[mfeurer]

Hi,
at the moment it is not possible to save the found models to a file. I will leave this issue open to remind us of adding this feature.

[mglowacki100]

Hi,
Here is nice approach to saving model:
http://rhiever.github.io/tpot/examples/IRIS_Example/

[mfeurer]

This is not the same as the thread creator asked for. You can achieve something similar by using AutoSklearnClassifier.show_models(), which prints you the members of the final ensemble, together with their weight. So far, this is only possible because of the rather simple ensemble considered and will possibly vanish in the near future.

[mfeurer]

auto-sklearn supports pickle with commit 9415f38

[devonbrackbill]

I'm not able to get pickle to work with commit 9415f38. When I install the development branch + the refactor/cleanupcomponents branch and run the test/test_automl/test_pickle.py file I still get the following error: TypeError: Pickling an AuthenticationString object is disallowed for security reasons. Any ideas what I'm doing wrong? Or how I could help make pickle work here?

[mfeurer]

Sorry for the late reply. I think there was an error on my side. Can you please try the latest version of the development branch? That should work. Is there any reason why you want to use refactor/cleanupcomonents?

[mfeurer]

Sorry, my bad. Of course you wanted to use the branch because the commit was in there. Please only use the development branch now.

[devonbrackbill]

Thank you! It works on the development branch. My mistake for using the incorrect branch.

[RedLiN5]

Hi guys. I've tried 9415f38 to save model.
But I got this error:

File "automl_classification.py", line 73, in main
    pickle.dump(automl, f)
_pickle.PicklingError: Can't pickle <class 'smac.runhistory.runhistory.RunKey'>: attribute lookup RunKey on smac.runhistory.runhistory failed

Does anyone know why it doesn't work?

[mfeurer]

See issue #138. A fix is under construction.

[namankumar]

A clean hack for saving the best model:

x = automl.show_models()
results = {"ensemble": x}
pickle.dump(results, open('fname.pickle','wb'))

[gstempeck]

When you save the models using show_models() like namankumar lists, how do you re-load and execute the predict() from the pickle file?

When I pickle.dump(classifierobj) I get a MemoryError but I assume that's the correct way to reload classifierobj to call predict().

[mfeurer]

Yes, that's the correct way to do this. I cannot explain when/why this happens as it never occured to me. Could you please provide some information about the data you're using (number of features and number of data points, sparse or not) and your system (especially RAM and bit)?