Python’s warnings module supports scoped suppression, allowing specific warnings to be silenced in a targeted block of code without affecting the rest of the application.

The warnings.catch_warnings() context manager temporarily overrides the active warning filters. Any filters added inside the block are discarded when it exits:

import warnings
import numpy as np

with warnings.catch_warnings():
    warnings.filterwarnings(
        "ignore",
        message="divide by zero",
        category=RuntimeWarning,
    )
    ratios = np.log(old_values / new_values)

ratios[~np.isfinite(ratios)] = 0.0

Targeted Filtering

Filters can be as broad or narrow as needed. The message parameter accepts a regex matched against the warning text, and category restricts it to a specific warning class. Both are optional — omitting them will match more broadly.

For cases where granularity isn’t important, simplefilter is a shorthand:

with warnings.catch_warnings():
    warnings.simplefilter("ignore")
    noisy_third_party_call()

Escalating to Errors

You can also use this pattern to strictly enforce zero warnings, which is useful in CI/CD environments or test suites:

with warnings.catch_warnings():
    # Turn all warnings into exceptions
    warnings.simplefilter("error")
    # This will now raise a RuntimeWarning exception
    call_deprecated_function()

Capturing Warnings

If you need to verify that a warning is issued (e.g., in unit tests), use record=True to capture them into a list instead of printing them:

with warnings.catch_warnings(record=True) as w:
    warnings.simplefilter("always") # Cause all warnings to always be triggered

    trigger_warning()

    assert len(w) == 1
    assert issubclass(w[-1].category, RuntimeWarning)
    assert "invalid value" in str(w[-1].message)