BUG: np.histogramdd loses precision on its inputs, leading to incorrect results #11023
Conversation
Previously a fuzzy rounded comparison was used for the rightmost bin of histogramdd. It's not clear why this was done, and it resulted in surprising behavior. This also removes the restriction that bin edges must be floats, and allows integer arrays to be passed (and returned) Fixes numpygh-10864
This is due to numpygh-11022.
| hist, edges = histogramdd((x, y), bins=(x_edges, y_edges)) | ||
|
|
||
| assert_equal(edges[0].dtype, x_edges.dtype) | ||
| assert_equal(edges[1].dtype, y_edges.dtype) |
There was a problem hiding this comment.
These fail on master, as edges[i].dtype is float64
| y = big + x | ||
| y_edges = big + x_edges | ||
|
|
||
| hist, edges = histogramdd((x, y), bins=(x_edges, y_edges)) |
There was a problem hiding this comment.
This also fails on master with
ValueError: `bins[0]` must be strictly increasing, when an array
There exist more malicious test cases that give the wrong result rather than crash, but I figured this was sufficient.
| y = np.array([0, 1, 2]) | ||
| x_edges = np.array([0, 2, 2]) | ||
| y_edges = 1 | ||
| hist, edges = histogramdd((x, y), bins=(x_edges, y_edges)) |
There was a problem hiding this comment.
On master, this test crashes with
ValueError: `bins[0]` must be strictly increasing, when an array
| @@ -646,7 +644,7 @@ def test_rightmost_binedge(self): | |||
| bins = [[0., 0.5, 1.0]] | |||
| hist, _ = histogramdd(x, bins=bins) | |||
| assert_(hist[0] == 0.0) | |||
| assert_(hist[1] == 1.) | |||
| assert_(hist[1] == 0.0) | |||
There was a problem hiding this comment.
This test was weird. 1.0000000001 <= 1.0 should be false.
|
Hmm, this does change behavior and might cause problems downstream, so it probably needs a compatibility note in the release notes. |
|
I'll add one. I don't think the change is substantially different from the rounding fixes added to |
This covers the changes made in numpygh-10324 and numpygh-11023
|
@charris: Added, along with |
|
Thanks Eric. |
Previously:
Note that this is different to gh-7845, which is about precision on the outputs.