fix: Tanhscaler nan output for constant feature

[ab93]

When the variance of input time series is zero, then we encounter a division by zero problem.
This led to NaN outputs.

This change checks if a feature is indistinguishable from a constant feature or not. If yes, then set the std as 1.0 for that feature. This will stop the feature from getting scaled by it's zero variance.

[codecov]

Codecov Report

Merging #153 (6882d08) into main (69006eb) will increase coverage by 0.25%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #153      +/-   ##
==========================================
+ Coverage   97.19%   97.45%   +0.25%     
==========================================
  Files          33       33              
  Lines        1213     1218       +5     
  Branches       89       89              
==========================================
+ Hits         1179     1187       +8     
+ Misses         26       24       -2     
+ Partials        8        7       -1     

Impacted Files	Coverage Δ
numalogic/preprocess/transformer.py	100.00% <100.00%> (ø)

... and 1 file with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[s0nicboOm]

Can we use Standard Scaler instead?

[ab93]

Can we use Standard Scaler instead?

@s0nicboOm as discussed, we should try extending the StandardScaler class and cover more edge-cases. Please open an issue so that we can keep a track of it.