mlpack - Python: LARS throws runtime error

[demoronator]

Issue description

Your environment

• version of mlpack: mlpack Python 4.0.0.post1, 4.0.0, 3.4.2
• operating system: Ubuntu 20.04.5 LTS

Steps to reproduce

Code:

train_x = [
    [0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
    [0.5, 0.5, 0.5, 0.5, 0.5, 0.5],
    [1.0, 1.0, 1.0, 1.0, 1.0, 1.0],
]
train_y_px = [0.0, 0.5, 1.0]
lars_px = mlpack.lars(input_=train_x, responses=train_y_px,
                      lambda1=0.1, lambda2=0.1,
                      verbose=True)

Result:

[FATAL] Number of responses must be equal to number of rows of X!

Expected behavior

The rows of X(input_) and responses are the same as three.
So it should return the dictionary without error.

Actual behavior

It throws a runtime error.

[rcurtin]

Indeed, your input looks reasonable and it should work. Out of curiosity, does the code succeed if you specify train_x in a transposed form? (Or train_y_px in a transposed form?)

[demoronator]

I tried

train_y_px = [
    [0],
    [0.5],
    [1],
]

It also failed with the same error.

[rcurtin]

Thanks for the report. It took a while to dig into it, but I found that the LARS method (and specifically LARS) happens to have a bug with how it is handled in the Python bindings, and some other languages too. If you want to read the details, check out #3327.

Until that PR is merged and released, you can work around the bug by transposing your X data matrix, like this:

train_x = [
    [0.0, 0.5, 1.0],
    [0.0, 0.5, 1.0],
    [0.0, 0.5, 1.0],
    [0.0, 0.5, 1.0],
    [0.0, 0.5, 1.0],
    [0.0, 0.5, 1.0],
]

mlpack 4.0.1 will be released soon (next couple weeks?) with a fix.

Thanks so much again for the report, and if you find more issues in the future please do not hesitate to report them! 👍

[demoronator]

Great. Thank you!

[rcurtin]

mlpack 4.0.1 is released now, so this should be solved. Thanks again for the report! 👍