firefly-cpp/NiaARM

Tests failing

zStupan opened this issue · 5 comments

zStupan commented

3 of the current tests are failing.

Full output: test_results.txt

firefly-cpp commented

Huh. I see. There are some floating point problems as well as other problems that appeared when feature names changed in a dataset.

@zStupan, can you please also manually (on the paper) evaluate whether supp, conf and fitness measures are correct? I believe we need additional tests.

zStupan commented

Will do.

firefly-cpp commented

Thanks @zStupan. We can also compare obtained results against uARMSolver results.

zStupan commented

With the Abalone dataset, thr solution is to convert the lists to numpy arrays and use self.assertTrue(numpy.allclose(minval, min_value)),...

With the Wiki dataset, the header in the csv is Feat1, Feat2, so the header variable in the test is wrong.

Now I'm gonna debug the support and confidence calculation, and compare the results to uARMSolver.

firefly-cpp commented

Awesome.

There exist also some variables (calculation of supp and conf) written in slovene language. Please examine.