Given that we offer exponential weighted average and std, and rolling count, avg, and std for update_by(), I think it's reasonable to support cum_count(), cum_avg(), and cum_std(). These are natural things to want to compute, and I'm having a tough time computing them by composing existing operations.
Here is the gap in cumulative/rolling operations compared to Numerics.java array operations:
count (cumulative, rolling already exists)
avg (cumulative, rolling already exists)
count_neg, count_pos, count_zero (can implement using modification to CumSpec)
Given that we offer exponential weighted average and std, and rolling count, avg, and std for
update_by()
, I think it's reasonable to supportcum_count()
,cum_avg()
, andcum_std()
. These are natural things to want to compute, and I'm having a tough time computing them by composing existing operations.Here is the gap in cumulative/rolling operations compared to
Numerics.java
array operations: