PyTwoWay v0.3.21

-Switch BLM to use variance instead of sds as weights

-Slight speedups to BLM and dynamic BLM

-Fix progress bar for BLM

PyTwoWay v0.3.20

-Switch from G'wG to groupby-sum in BLM (approximately 2x faster overall than before)

-Add option to compute certain operations using GPU for BLM

PyTwoWay v0.3.17

-Move CHK event studies from BipartitePandas into PyTwoWay

-Add option to compute standard errors and clustered standard errors for continuous control variables in AKM estimation

-Fix bug with Jacobi preconditioner if any diagonal elements equalled 0

-Change parameter dictionaries to import from ParamsDict rather than BipartitePandas

-Add dynamic BLM (not yet added to documentation)

-Add interacted BLM (not yet added to documentation)

-Other small changes and bug fixes

PyTwoWay v0.3.15

-Fix FEControlEstimator to normalize categorical variables correctly (normalize the first type for each categorical variable)

-Change FEEstimator and FEControlEstimator to normalize the first firm rather than the last firm

Thank you to @Alalalalaki for pointing out the error with normalizing categorical variables and to @santiagohermo for suggesting to use releases!