Will Townes
Will Townes
Currently this is only supported for minibatch='none'. See if it could work with minibatch='stochastic' or 'memoized' as well. Relates to #15 because the estimation method will probably need to be...
GLM-PCA should support DelayedArray, HDF5Array, TENxMatrix, and tiledb (both sparse and dense) objects with memoized and stochastic gradient methods.
As a GLM-PCA user with covariates such as batch labels, I can learn how to adjust for these by creating a design matrix and passing it to the GLM-PCA functions....
It would be straightforward to include the Gaussian family, which would allow an alternative algorithm for computing PCA. In particular this may help scale PCA to out-of-memory datasets when we...
To help users pre-filter features and remove those that are not informative (so as to speed computation of glmpca), provide the function to compute deviance for each feature and rank...
Consider changing the packaging to the [poetry](https://python-poetry.org/) system.
I run my analysis in small pipelined chunks of commands on a server. This means that if there is an error, all the sequence of commands I am running will...
It would be straightforward to include the Gaussian family, which would allow an alternative algorithm for computing PCA. In particular this may help scale PCA to out-of-memory datasets when we...
Currently (due to its R origin) the glmpca method is implemented with frequent indexing along columns of various matrices. This is because R stores data in column-major order so it...
This looks like a useful tool! Have you considered making it into a package so folks can install from something like pypi?