Closed pzivich closed 2 years ago
In v0.4 there are also the following additional changes: re-organizing the internal structure for built-in estimating equations (no changes on the user-side though), more general functions for regression (along with some depreciation of functions), and adding further support for penalized regression.
Primary addition is better error handling as mentioned in #8
Previous versions sometimes failed silently. If more
init
's were provided than rows returned bystacked_equations
, then solvers likenewton
would return values of approximately zero for everything. Therefore, an answer would be returned but in actually the root-finding just failed to improve.This pull request addresses those errors by making sure there is no mismatch. Notice that I don't cover all mistakes with informative errors. For example, some mistakes will still be caught by
IndexError
's or when a 2d object and 1d object are incorrectly multiplied together. These will raise other errors