MRG: compatibility with numpy 1.12.0 #419
Conversation
Numpy 1.12 more strict about enforcing integers for indices.
Fix errors for Numpy 1.12 and floating point values used as indices.
Make indices integers; make boolean index match size of indexed array.
Make general function for multiplying a sequence of numbers. Use to return integer from product of array shape.
Enforce integer indices in kernel smooth indexing.
Make sure indices are integers rather than floats or arrays.
Force shape elements to ints for indexing.
Indexing now requires integers; force indices to ints, replicating previous behavior.
math.floor returns an integer.
The row selection for these tests was broken, depending on a single False for an empty array. The rows were being selected by comparison to an int, but the code changed to produce character strings some time back, so no rows were being found.
Add doctest flag to test equality of arrays reqpresented by their reprs in docstrings. This also fixes a change in the display of integers as floats. It appears that numpy 1.12 displays "1." compared to previous numpy display of "1.0".
Make _slicer a method rather than a lambda, and expand names for clarity. On the way, ensure slice indices are ints, not floats.
Make new slice copies at each iteration to prevent later iterations overwriting the earlier slices in memory. Extend tests.
A backport from a fix in nibabel.
math.ceil, math.floor, round generate integers in Python 3 but not Python 2.
|
Changes Unknown when pulling 471da90 on matthew-brett:np-12-compat into ** on nipy:master**. |
To allow for unresolved mroi test failure - nipy#420.
Current coverage is 83.18% (diff: 89.06%)
|
|
Changes Unknown when pulling a94a998 on matthew-brett:np-12-compat into ** on nipy:master**. |
| @@ -234,7 +236,7 @@ def cross_validated_update(self, x, z, plike, kfold=10): | |||
| if np.isscalar(kfold): | |||
| aux = np.argsort(np.random.rand(n_samples)) | |||
| idx = - np.ones(n_samples).astype(np.int) | |||
| j = np.ceil(n_samples / kfold) | |||
| j = int(math.ceil(n_samples / kfold)) | |||
There was a problem hiding this comment.
Shouldn't n_samples / kfold already be an integer (at least in Python 2)? Or does from __future__ import division not need to happen in every file where it's used?
There was a problem hiding this comment.
Yes, will be an integer in Python 2. I guess the question for @bthirion is - was this intended to be true (floating point division) even in Python 2? At the moment this can give different answers on Python 2 and Python 3.
There was a problem hiding this comment.
Yes, it is an int. I think that the result is equivalent in Python 2 and 3 ?
There was a problem hiding this comment.
If kfold doesn't divide evenly, it will be off by one:
Suppose n_samples = 5, kfold = 2.
Python 2: int(math.ceil(5 / 2)) == int(math.ceil(2)) == int(2.0) == 2
Python 3: int(math.ceil(5 / 2)) == int(math.ceil(2.5)) = int(3.0) == 3
There was a problem hiding this comment.
All right. In that case, we don't really care about the exact value, so both are fine.
| @@ -157,7 +159,7 @@ def ward_test_more(n=100, k=5, verbose=0): | |||
| # Check that two implementations give the same result | |||
| np.random.seed(0) | |||
| X = randn(n,2) | |||
| X[:np.ceil(n/3)] += 5 | |||
| X[:int(math.ceil(n / 3))] += 5 | |||
There was a problem hiding this comment.
@bthirion - how about this one? Presumably meant to be true division?
There was a problem hiding this comment.
@bthirion - I was thinking you meant this be floating point division, otherwise the ceil call doesn't make sense, because the integer division does an implicit floor on this positive number. But you did mean integer division? In which case we can discard the ceil.
Use true division / math.ceil to find indices.
|
Changes Unknown when pulling b148483 on matthew-brett:np-12-compat into ** on nipy:master**. |
|
Thanks for the review and comments. |
Numpy 1.12 started raising errors for float indices in a variety of
circumstances, causing lots of test errors - e.g.
https://travis-ci.org/nipy/nipy/jobs/193952659