pairwise_tests producing array of t values internally

[George3d6]

I occasionally get mysterious crashes when using the pairwise t-test, e.g:

  .../pingouin/pairwise.py", line 397, in pairwise_tests
    df_ttest = ttest(
  .../pingouin/parametric.py", line 310, in ttest
    bf = bayesfactor_ttest(tval, nx, ny, paired=paired, alternative=alternative, r=r)
  .../pingouin/bayesian.py", line 128, in bayesfactor_ttest
    assert isinstance(t, (int, float)), "The T-value must be a int or a float."
AssertionError: The T-value must be a int or a float.

This happens when using a fairly innocuous dataframe:

                                between                                dv
0                              2.0                                         4.00
1                              1.0                                         7.00
2                              3.0                                        13.00
3                              2.0                                         8.00
4                              2.0                                         6.00
5                              2.0                                         5.50
6                              1.0                                        10.00
7                              1.0                                        10.00
8                              2.0                                        10.00
9                              1.0                                         6.00
10                             2.0                                         9.00
11                             1.0                                        13.00
12                             2.0                                        12.00
13                             2.0                                        12.00
14                             1.0                                         7.75
15                             1.0                                         5.50
16                             1.0                                         6.00
17                             2.0                                         9.00
18                             1.0                                         8.50

The issue seems to arise from the fact that the t variable being checked by the assert is actually a single-element array (e.g. [0.521])

I've been unable to figure out the exact cause of the bug or replicate it, any help with this would be appreciated.

[George3d6]

I've implemented a potential fix here: #299

[George3d6]

@raphaelvallat can I ask why this is marked as invalid?
Looking at the code (see linked PR) there's a clear issue with how the scipy interface is used, and this is causing issues in certain cases, since a tuple is treated as if it were a float/int

[raphaelvallat]

@George3d6 --> Invalid does not mean that the issue is not valid, but rather that something doesn't look right in the Pingouin code. Apologies for the confusion. I'll update the label to "bug".

[George3d6]

@raphaelvallat as far as I can see this isn't a recent change in the scipy interface as I suspected, so I guess the bug was always there.