Implement moving histogram algorithm

[gavv]

Add a class that implements moving histogram.

The idea is simple:

• a few parameters are given: value range, number of bins to split range into (N), and value window length (K)
• we allocate N bins, each one corresponds to sub-range of value range
• we allocate a ring buffer for last K values
• each bin holds a counter
• when a new value is added, we:
  • add it to ring buffer tail
  • remove oldest value from ring buffer head
  • increment counter of the bin corresponding to the added value
  • decrement counter of the bin corresponding to the removed value

The new class can be named MovHistogram. Interface should be similar to core::MovStats class, which implements moving minimum, maximum, and variance.

We need a method to add a value, and to get counter value for every bin.

We also need unit tests, you can find example here.

Please use core::RingQueue for ring buffer and core::Array for bins.

This task is needed for #712.

[spran180]

Hi, I am interested in solving this issue.

[gavv]

@spran180 Welcome, thanks!

[gavv]

What to cover in unit tests, as discussed in #760:

• T = float
• min and max < 0
• win_length == 1
• when there are multiple values in every bin (currently every bin is either 0 or 1)
• how bin counters increment and decrement when rolling window is moved (e.g. you call add_value, some bin changes from N to N-1, and another bin changes from K to K+1)
• out of bounds values (clamping)

[spran180]

Hey @gavv,

I'll soon create the pull request including these tests.

[gavv]

Landed