Rolling argmax and argmin functions

[testforvln]

Could you add rolling argmax and argmin functions? Just to return the position of the max/min value.

[AdilZouitine]

Hello,
Yes, it is possible, but I don't see the point of having a rolling argmin?
Do you have an example? 😄

[testforvln]

Yeah, when calculating some factors in quant finance, the rolling argmin/argmax functions are very useful.

[AdilZouitine]

I understand the interest!
Do you want to make the pull request? If not, I can do it next weekend after my holidays 😄

I will also move the repository to https://github.com/online-ml and probably rename the library.

[testforvln]

Sorry that I failed to implement argmin/argmax functions by myself. It would be very nice of you to implement these functions anytime you like.

[AdilZouitine]

No problem 😄 I'll do it!

I'll ping you when it's done.

[testforvln]

Hi, do you have any progress on this issue? I'm wondering if the argmin/argmax function could only be implemented by binary search the item each time on updating.

[AdilZouitine]

Hey, I'm on vacation; I add this issue to my todo list 😄

[testforvln]

Sorry to bother you. Have a good time!

[AdilZouitine]

Hi @testforvln,
I implemented the rolling argmin; the code is here: https://github.com/online-ml/watermill.rs/blob/rolling_argmin/src/argmin.rs
I made some caching to avoid searching the argmin at each update.

[testforvln]

That's great! I thought of maintaining another sorted order number for sorted_window, but every update would cost O(n) for maintaining the sequence. Your solution of caching is very suitable for this argmin function.