Warning that seems a bug in Matthews correlation coefficient computation

Question

The grid search part of my code looks like this:

svc_param_grid={"C":c, "kernel":kernel, "gamma":gamma, "degree":degree}
grid_cv_object = GridSearchCV(
    estimator=SVC(),
    param_grid=svc_param_grid,
    n_jobs=-1,
    verbose=3,
    cv=k_fold,
    # a callable that returns a single value
    scoring=make_scorer(matthews_corrcoef)
)

It is used for multi-class classification (predicting protein secondary structure sequence).

Why am I getting this error and is there a workaround?

/usr/local/anaconda3/lib/python3.8/site-packages/sklearn/metrics/_classification.py:873: RuntimeWarning: invalid value encountered in double_scalars
  mcc = cov_ytyp / np.sqrt(cov_ytyt * cov_ypyp)

Answer 1

You are experiencing this issue, which was fixed in this pull request, after the 0.24.2 release.

The warning is caused by a division-by-zero due to no variance in your predictions. It is misleading and can be ignored; the metric is correctly returning zero.