accuracy_score and clf.evaluate of Autokeras are not the same

[Riccardo Zuliani]

Dataframe tested: 10.csv

[Riccardo Zuliani]

I find out that is a problem with the if not isinstance(df, pd.DataFrame): in line 118.
In fact if i want to run a test the if will be true, on the other hand if I want to run an OpenML Benchmark the if will false.

The problem is that return_X_y(df) take in input a DataFrame in case of OpenML Benchmark or a couple of DataFrame in case of Kaggle Benchmark

[Riccardo Zuliani]

I'll re-check what input type the all algorithms accept to make sure every dataframe is adapt on the usage of that specific algorithm

[Riccardo Zuliani]

Implement with the copy library a deep copy of the dataframe before every algorithm manipulation