chengfgao / TopicVelo

TopicVelo: Dissection and Integration of Bursty Transcriptional Dynamics for Complex Systems
BSD 3-Clause "New" or "Revised" License
19 stars 2 forks source link

Transition matrix integration #5

Open hdanderson opened 7 months ago

hdanderson commented 7 months ago

Currently, integrating the global transition matrix is either very runtime intensive or memory intensive. I can run the memory intensive version for 20,000 cells on my 32 GB RAM Mac or with an 80 GB RAM job on a compute core, but this is not ideal. We need to fix this by using smarter matrix multiplication and sparse array memory allocation that calculates nonzero elements prior to initialization.