Implement LArray.reverse(axis)

[gdementen]

So that we can write:

(agei * dx).reverse(AGE).cumsum(AGE).reverse(AGE)

instead of the much less readable:

(agei * dx)[AGE[::-1]].cumsum(AGE)[AGE[::-1]]

[alixdamman]

My biggest fear with small methods like that is to run into the same pitfall as pandas: having a too wide API and then make users to use only (less than) 10% of it.
I wonder if we shouln't encourage users to have an utils.py module in their models instead.

[gdementen]

Pandas has indeed a very wide API, but I think it is part of the reason it is so successful. Yes, each user only ever uses 10% of Pandas, but not the same 10% and the long tail of features is important.

Pro arguments

• the slice step is very seldom meaningful in the context of labelled arrays so (non Python developers) users do not really know this syntax.
• the number one priority of LArray is to make code readable and reverse is IMO much more readable than [::-1].
• reverse is probably more easily discoverable than [::-1] for our typical user base. If one of our current users is looking for a way to reverse an axis, there are chances he finds the method. The inverse slice? I don't think so.

Con arguments

• it adds more code to maintain. That's a true problem, but I think there are other places where we have more to gain in this regard: e.g. think about how to have less ways to call the functions which accept 23 different types as arguments.
• that is two ways to do the same thing and we would promote the one which is not builtin/idiomatic Python.

All in all, I would say the benefits far outweigh the disadvantages for our non-Python expert users. For seasoned Python users, the status quo is clearly better IMO, but this is not our target audience at the moment.