Add epsilon to MTER progress postfix to prevent ZeroDivisionError

[Shxiao101]

Description

MTER.fit can still raise ZeroDivisionError when every sampled BPR pair is skipped, because the "correct" percentage in the progress postfix divides by n_bpr_samples - skipped.

#677 and #686 added + 1e-8 to this denominator in BPR, WBPR, Companion, ComparERSub and LRPPM, but cornac/models/mter/recom_mter.pyx was not included. This PR applies the same one-line change to MTER.

Note that the postfix is computed even with verbose=False (only the bar display is disabled), so the error also happens in non-verbose runs.

Reproduction (2 users x 2 items, all rated 5.0, so every sampled pair is skipped):

from cornac.data import SentimentModality
from cornac.eval_methods import BaseMethod
from cornac.models import MTER

ratings = [("u1", "i1", 5.0), ("u1", "i2", 5.0), ("u2", "i1", 5.0), ("u2", "i2", 5.0)]
sentiment = [
    ("u1", "i1", [("a1", "o1", 1.0)]), ("u1", "i2", [("a2", "o2", 1.0)]),
    ("u2", "i1", [("a1", "o2", -1.0)]), ("u2", "i2", [("a2", "o1", 1.0)]),
]
bm = BaseMethod.from_splits(train_data=ratings, test_data=ratings, exclude_unknowns=True,
                            sentiment=SentimentModality(data=sentiment), seed=123)
MTER(n_user_factors=2, n_item_factors=2, n_aspect_factors=2, n_opinion_factors=2,
     n_bpr_samples=10, n_element_samples=4, max_iter=3, n_threads=1, seed=123).fit(bm.train_set)

Before this change:

  File "cornac/models/mter/recom_mter.pyx", line 422, in cornac.models.mter.recom_mter.MTER.fit
ZeroDivisionError: float division by zero

After this change, fit completes.

Testing (Windows, Python 3.12, MSVC build via python setup.py build_ext --inplace):

• Reproduction above: fails on master, passes with this change.
• flake8 . --select=E9,F63,F7,F82 (same as CI): 0.
• pytest tests/cornac/models tests/cornac/data tests/cornac/eval_methods: 139 passed.

Related Issues

Follow-up to #675 (fixed for BPR in #677 and for other BPR-based models in #686).

Checklist:

• I have added tests. (not added; Fix ZeroDivisionError on BPR by adding an epsilon #677 and fix: add epsilon to prevent ZeroDivisionError in BPR-based models #686 made the same change without tests)
• I have updated the documentation accordingly. (N/A)
• I have updated README.md (if you are adding a new model). (N/A)
• I have updated examples/README.md (if you are adding a new example). (N/A)
• I have updated datasets/README.md (if you are adding a new dataset). (N/A)

🤖 Generated with Claude Code

[qtuantruong]

Thanks for the fix!