Add array_emd function #26
Conversation
There was a problem hiding this comment.
Thanks for the PR!
Seems like there's pretty bad precision in certain cases, like array_emd([1], [2]). This should be 1, but it's 0.9666... on my machine. This has to do with NumPy's heuristics for bin sizes, which result in many bins even in small cases like that. I'm not sure how important this is, but it may be worth thinking about.
pyemd/emd.pyx
Outdated
| histogram. Defaults to the range of the union of `first_array` | ||
| and `second_array`. Note: if the given range is not a superset | ||
| of the default range, no warning will be given. | ||
|
|
There was a problem hiding this comment.
In general, trailing whitespace should be stripped. There's probably a plugin for your text editor that can do this automatically for you.
pyemd/emd.pyx
Outdated
| @@ -45,6 +46,10 @@ def validate(first_histogram, second_histogram, distance_matrix): | |||
| if (first_histogram.shape[0] != second_histogram.shape[0]): | |||
| raise ValueError('Histogram lengths must be equal') | |||
|
|
|||
| def euclidean_pairwise_distance(x): | |||
There was a problem hiding this comment.
There should be two blank lines before top-level function definitions. There are automated tools for checking these sorts of things, called “linters”, such as pylint.
pyemd/emd.pyx
Outdated
| return emd(first_histogram, | ||
| second_histogram, | ||
| distance_matrix, | ||
| extra_mass_penalty) |
There was a problem hiding this comment.
This call should explicitly pass extra_mass_penalty as a keyword argument:
emd(..., extra_mass_penalty=extra_mass_penalty)
README.rst
Outdated
| >>> first_array = [1,2,3,4] | ||
| >>> second_array = [2,3,4,5] | ||
| >>> array_emd(first_array, second_array, bins=2) | ||
| 0.5 |
There was a problem hiding this comment.
There should be a few unit tests added to test/test_pyemd.py, in addition to this.
pyemd/emd.pyx
Outdated
| @@ -91,6 +96,76 @@ def emd(np.ndarray[np.float64_t, ndim=1, mode="c"] first_histogram, | |||
| distance_matrix, | |||
| extra_mass_penalty) | |||
|
|
|||
| def array_emd(first_array, | |||
There was a problem hiding this comment.
I think this should be called emd_samples (if you can think of something better, I'm all ears), since the other emd functions also take arrays, and they all begin with emd.
pyemd/emd.pyx
Outdated
| max(np.max(first_array), np.max(second_array))) | ||
|
|
||
| if type(bins) == str: |
There was a problem hiding this comment.
isinstance(bins, str) should be preferred over explicitly checking type equality.
pyemd/emd.pyx
Outdated
| if type(bins) == str: | ||
| hist, _ = np.histogram(np.concatenate([first_array, | ||
| second_array]), | ||
| range=range, bins=bins) |
There was a problem hiding this comment.
Nitpick: arguments should be aligned.
pyemd/__init__.py
Outdated
| @@ -33,6 +33,14 @@ | |||
| >>> emd_with_flow(first_signature, second_signature, distance_matrix) | |||
| (3.5, [[0.0, 0.0], [0.0, 1.0]]) | |||
|
|
|||
| You can also calculate the EMD directly from two arrays: | |||
There was a problem hiding this comment.
Could elaborate and say “arrays of observations” or “samples” instead of “arrays” and mention that the histogram is computed automatically.
pyemd/emd.pyx
Outdated
| if distance == 'euclidean': | ||
| distance = euclidean_pairwise_distance | ||
|
|
There was a problem hiding this comment.
Nitpick: comments should be capitalized, to be consistent with the others.
|
Apologies, I didn’t intend for this to be done – I just need to push to go between my windows and Linux environments. Thanks for your comments – I’ll add a comment to the PR when I think it’s done.
|
|
Gotcha, no worries. I'll wait to review until you comment. |
|
Okay, I'm pretty happy with that now. Thanks for your patience! |
|
Great, looks good. Thank you! |
Release v0.5.0 - Add the `emd_samples()` function (PR #26). - Clarify docstrings. - Update documentation in README. - Refactor tests. - Explicitly support Python 3.4 and 3.5.
|
Released in v0.5.0. |
As discussed in #25