Errors with behavioral and mean-centered PLS

[tylercall]

Hello,

I recently encountered errors while trying to run behavioral and mean-centered PLS on my data (functional connectivity matrix and accompanying behavioral matrix). I have been able to reproduce these errors with the sample code from the README, leading me to believe the issue might be a bug in in the PLS implementation and not an issue with my input matrices (which I've confirmed are correctly stacked and contain no constant columns).

For both PLS functions, I tried testing different combinations of arguments (e.g., omitting groups, setting it to [40] to represent a single group, etc.). The only way I have been able to bypass the errors is to turn off permutation testing, bootstrapping, and force n_proc=1.

Behavioral PLS

For behavioral PLS, the error is: "ValueError: matmul: Input operand 1 has a mismatch in its core dimension 0, with gufunc signature (n?,k),(k,m?)->(n?,m?) (size 4 is different from 40)". I have reproduced the error with the following sample code:

    X = np.random.rand(80, 10000)  # a 10000-feature (e.g., neural) data array
    Y = np.random.rand(80, 10)     # a 10-feature (e.g., behavioral) data array
    groups = [20, 20]  # a list with the number of subjects in each group
    n_cond = 2         # the number of tasks or conditions

    bpls = pyls.behavioral_pls(X, Y, groups=groups, n_cond=n_cond)

Mean-centered PLS

For mean-centered PLS, I am getting "IndexError: too many indices for array: array is 1-dimensional, but 2 were indexed". This error occurs with the following sample code:

    X = np.random.rand(80, 10000)  # a 10000-feature (e.g., neural) data array
    groups = [20, 20]  # a list with the number of subjects in each group
    n_cond = 2         # the number of tasks or conditions

    mpls = pyls.meancentered_pls(X,
                                 groups=groups,
                                 n_cond=n_cond)

"Workaround"

The only way I've been able to bypass this error is by forcing n_boot=0, n_perm=0, and n_proc=1. For example:

mpls = pyls.meancentered_pls(X, groups=groups, n_cond=n_cond, n_boot=0, n_perm=0, n_proc=1)

Running without permutation testing or bootstrapping is not really a viable workaround, but I hope that this helps pinpoint the bug (or user error?).

---

Thank you!

[tylercall]

Update: I have confirmed that these errors do not occur with the previous release (0.0.1), either with the sample data/code or with my data/code

[liuzhenqi77]

Hi,

We are working in #70 to solve this, meanwhile, please revert the last two commits with:

pip install git+https://github.com/netneurolab/pypyls.git@ee5e3e03df4b7fd3ed83f3a858991cc5672d0be5

Thanks.