Previously (#117), through a sweep of different z dimensions, I saved a single compressed matrix for each z dimension even though I was training multiple. In #117 I also calculated stability measurements (correlation matrix determinants) - however, I was getting underflow as the dimensionality increased.
A solution is to save the population of models that are being trained. Saving the ensemble of models for each z will enable downstream stability estimates and also potentially improve biological discovery in Tybalt and ADAGE models. For convenience, I will call these models ezADAGE and ezTybalt :joy_cat:
I also update the parameter file (as a result of ADAGE tied weights in #123 ) (also related to #127 )
Previously (#117), through a sweep of different z dimensions, I saved a single compressed matrix for each z dimension even though I was training multiple. In #117 I also calculated stability measurements (correlation matrix determinants) - however, I was getting underflow as the dimensionality increased.
A solution is to save the population of models that are being trained. Saving the ensemble of models for each z will enable downstream stability estimates and also potentially improve biological discovery in Tybalt and ADAGE models. For convenience, I will call these models
ezADAGE
andezTybalt
:joy_cat:I also update the parameter file (as a result of ADAGE tied weights in #123 ) (also related to #127 )