Wildboar ShapeletForestClassifier not working? about wildboar HOT 7 CLOSED

Hi!

Thanks for your bug report!

The example in the tutorial seems to be a little unintuitive since I load two datasets (synthetic_control and two_lead_ecg) and define x_train etc two times. The actual output is from the not where the training and testing data is split from the full data (i.e., from synthetic_control) (which contains the classes 1-6, thus 6 is a possible outcome). In the code you submit as part of this issue the two_lead_ecg dataset is used which only contains the classes 1 and 2 so the output of your example actually make sense.

If i run your example the resulting model has an accuracy of 90% and if I change the metric to "scaled_euclidean" the accuracy the expected 96% accuracy.

As for the other examples the accuracy i get using the following snippet:

def test_dataset(dataset):
    x_train, x_test, y_train, y_test = load_dataset(dataset, merge_train_test=False)
    clf =ShapeletForestClassifier(metric="scaled_euclidean", random_state=123, n_jobs=-1)
    clf.fit(x_train, y_train)
    return clf.score(x_test, y_test)

These are the results I get:

>>> test_dataset("GunPoint")
0.9866666666666667
>>> test_dataset("CricketX")
0.764102564102564
>>> test_dataset("CBF")
0.9722222222222222

To fully resolve the issue could you tell me your:

• Operating system
• Wildboar version
• Python version

Thanks again your bug report!

from wildboar.

Hello,
Thanks for your response. I've attached a screenshot of what I get when I run your code above - tl;dr its still giving pretty poor accuracy (49.3%).

My Operating system is Windows 10 20H2, Wildboar version is 1.0.9 and Python version is 3.7.10. Incidentally I'm working in a miniconda environment. However I tried this on my native/non-anaconda Python environment (also Python 3.7.9) and had the same problem.

from wildboar.

Thanks for your report!

I will try to debug the issue once I get hold of a windows computer.

Could you try with at 64bit installation of miniconda?

from wildboar.

Hello,
Thanks for your prompt response!
Sadly, no - I've only the one computer. But, the other Python installation I told you about (i.e the non-Anaconda one I also have) is 64-bit, and shows the same issue..

from wildboar.

The issue seems to affect only Windows. Pre-build binaries and code compiled from source.

from wildboar.

Fixed in version 1.0.10.

from wildboar.

Confirmed fixed, thanks! Got 1.0.10 from pip. Thanks again :)

from wildboar.