v1.0.6

• lags creation of covariates reviewed
• effective data casting to float32, the subsetting operation was recasting data in float64, fixed
• various speed-ups

v1.0.5

• preimpution of small gaps is computed once and shared to all columns imputations -> speed-up when subset-cols is not a single column
• redundancy in columns passed to the estimator in the find_best_lags methods -> speed-up

v1.0.4

• typing: python 3.8 compatibility

v1.0.3

• Extended suite tests
• Float32 casting before entering the imputation engine

v1.0.2

• sample_weights for FastRidge

v1.0.1

• numerous bugs fixes

v1.0

timefiller 1.0

• Default estimator to custom implementation of Ridge regression ; no scikit-learn overhead for speed
• multivariate_lags is 'auto' by default ; it seeks for optimal lags in the covariates during the
  imputation of each column
• Numerous speed-ups improvements

v0.1.10

Version 0.1.10 - Release Notes

Enhancements

• Code Refactoring:
  • Improved functionality around the multivariate_lags argument.
  • Optimal forward and backward lag selection for covariates is now performed for each imputed feature.
  • Accelerated computation of cross-correlations using Numba, with robust handling of NaN values.

Documentation

• Docstrings Added:
  • Comprehensive docstrings have been included to enhance code clarity and usability.