Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

leave_k_out_split produces incorrect values in test set #639

Open
chrisjkuch opened this issue Dec 23, 2022 · 1 comment · May be fixed by #640
Open

leave_k_out_split produces incorrect values in test set #639

chrisjkuch opened this issue Dec 23, 2022 · 1 comment · May be fixed by #640

Comments

@chrisjkuch
Copy link

chrisjkuch commented Dec 23, 2022 •
edited
Loading

I noticed odd behavior in leave_k_out_split - specifically, under certain circumstances (many rows with one value?), the number returned for the withheld test set value is different from the actual value. This causes train + test to fail to reconstruct the original matrix.

Script to reproduce:

import implicit

implicit.__version__
#> '0.6.2'

from implicit.evaluation import leave_k_out_split
from scipy import sparse

ratings = sparse.csr_matrix(
    [[3, 2, 1, 0], [1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 1, 1, 1]]
)

train, test = leave_k_out_split(ratings, K=1, random_state=42)

diff = (train + test) - ratings

diff.nnz
#> 1

train.todense()
#> matrix([[3, 2, 0, 0],
#>         [1, 0, 0, 0],
#>         [0, 1, 0, 0],
#>         [0, 0, 1, 0],
#>         [0, 1, 0, 1]])

test.todense()
#> matrix([[0, 0, 4, 0],
#>         [0, 0, 0, 0],
#>         [0, 0, 0, 0],
#>         [0, 0, 0, 0],
#>         [0, 0, 1, 0]])

diff.todense()
#> matrix([[0, 0, 3, 0],
#>         [0, 0, 0, 0],
#>         [0, 0, 0, 0],
#>         [0, 0, 0, 0],
#>         [0, 0, 0, 0]])

Created at 2022-12-23 13:59:28 CST by reprexlite v0.5.0

I believe the issue is in _take_tails - the returned test_idx array has multiple copies of the first user index returned, so we end up with a test set value that is copied multiple times.

(As an aside, the call to _take_tails when shuffled=True does not pass on the rng, so the random state cannot be maintained.)
@chrisjkuch chrisjkuch linked a pull request Dec 23, 2022 that will close this issue
Open
@chrisjkuch
Copy link
Author

Existing test can also be made to fail occasionally if density is set below 0.1:

implicit/tests/evaluation_test.py

Line 18 in 871e0c7

mat = random(100, 100, density=0.5, format="csr", dtype=np.float32)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Fix incorrect test set values in leave_k_out splits with sparse user rows chrisjkuch/implicit
1 participant
@chrisjkuch