Triangle addition fails for triangles with the same columns, but different ordering.

[genedan]

Are you on the latest chainladder version?

• Yes, this bug occurs on the latest version.

Describe the bug in words

Discovered while working on #666. When you reorder a triangle's columns, you can't add it back to another triangle that is the same in every other way except for the ordering. This used to be a valid operation, since it's taken from the tutorial.

How can the bug be reproduced?

import chainladder as cl
s1 = cl.load_sample('clrd').iloc[:3]
s2 = s1.sort_index(ascending=False)
s3 = s1.iloc[:, ::-1]

s1 + s3

What is the expected behavior?

The result should be:

np.True_

[jbogaardt-wcf]

I agree, this used to work per your expectations. Triangle arithmetic tries to take the best from both numpy and pandas.
Numpy doesn't care about axis labels, so arithmetic will work when shapes match OR when broadcasting is appropriate.
Pandas does label matching, but doesn't really support broadcasting, so the following operation isn't really helpful in pandas, but we try to make it helpful in chainladder by allowing broadcasting to take precedence:
loss_ratios = triangle[['paid', 'incurred']] / triangle['premium']

In the test example above, broadcasting is not in play, therefore label matching should be enforced (across all dimensions) but it doesn't.