ENH: Add support for calculating EWMA with a time component #34839
Conversation
|
Hello @mroeschke! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found: There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻 Comment last updated at 2020-07-06 21:33:14 UTC |
mroeschke
changed the title
|
|
||
| last_result = vals[0] | ||
|
|
||
| for i in range(N): |
There was a problem hiding this comment.
can you you put this in a nogil block? do we have timings on this?
There was a problem hiding this comment.
I tried to nogil this but could not since we are doing array arithmetic & broadcasting.
weights = 0.5 ** ((times[i] - times[mask.view(np.bool_)]) / halflife)
observations = vals[mask.view(np.bool_)]
last_result = np.sum(weights * observations) / np.sum(weights)
I can add an ASV for this.
There was a problem hiding this comment.
ok sure on the asv.
I think you can convert the mask view to a np.take operation (but can always just add an issue for this) if needed
pandas/tests/window/test_ewm.py
Outdated
| tm.assert_frame_equal(result, expected) | ||
|
|
||
|
|
||
| def test_ewma_with_times_variable_spacing(): |
There was a problem hiding this comment.
can you add an example that has times in DTI w/tz (I think answer should be the same as naive but just checking)
|
cc @pandas-dev/pandas-core if anyone has comments. |
|
thanks @mroeschke |
black pandasgit diff upstream/master -u -- "*.py" | flake8 --diffThis enhancement allows EWMA to be calculated relative to the timestamp in which an observation occurs instead of assuming each observation is equally spaced.