Releases: leonidgarin/TIDE
Releases · leonidgarin/TIDE
Release list
v0.5.0
- Fixed bug with incorrect behaviour in TIDE algorithm: if candidate point was idx_base, it was skipped.
- Changed logic for point selection and brackets.
Now for positive trend we use the candidate point (idx_cand) and (a,b] brackets; for negative - idx_cand+1 and [a,b) brackets - Changed text brackets for -inf, +inf borders to always be '(' or ')'
- Visual improvements for plots
- Added sample_rate in stats for convenience
- Rewrited some docs
- Added 'tide' mode for categorical bins handling
v0.4.0
- Add parameter 'handle_unknown' to 'TIDE.transform' which specifies behavior on how to transform values that are not covered by fitted bins
- One more attempt to fix PSI calculation when there are no continuous values in column
v0.3.4
- Add basic tests for event rate calculation, as well as WoE, IV, PSI
- Fix bug of PSI calculation when continuous index not provided explicitly
v0.3.3
- Fix PSI calculation when period column has string format
- Fix reset functionality
v0.3.2
- Change PSI to PSI_base and PSI_seq in stats.
PSI_base compares periods with first one, while PSI_seq does period by period sequentially (which was the initial PSI realization) - Plots improvement: add IV to header, add new PSIs.
v0.3.1
- Fix case when the missing_strategy is "worst_bin" or "best_bin", but there is no missing values present in column
v0.3.0
- Add missing value merge to worst/best continuous bins
v0.2.1
- Fix chi-merge for tiny categorical bins
- Add warning for columns with >= 100 categorical bins. As chi-merge for categorical bins has O(n^2) complexity, performance may be poor in case when one chooses cat_strategy='chi-squared'
v0.2.0
- Add Chi-merge for categorical bins
- Add min_er_delta which forces the minumum difference between event rates in bins across all periods
- Change min_sample_rate logic: now the denominator corresponds to the size of full x column (previously it was of continuous part only)
- Minor fixes and improvements
v0.1.2
- Fix issue with incorrect IV calculation for fully categorical variables
- Remove unnecessary debug prints