Can Ergen

icon company University of California Berkeley icon location Berkeley icon location Computational Biology PostDoc at UC Berkeley and Weizman Institute of Science Clinician Scientist at University Hamburg

Results 98 comments of Can Ergen

Using torch.compile in Pyro models

I tried it out on my side and got some cryptic error messages (it was on a private repo with a not published model though). My idea was to call...

Using torch.compile in Pyro models

Yes, that's my understanding of how we do guide warmups for Pyro (e.g. during loading a trained model). I don't think pyro.clear_param_store() is necessary here.

question on mismatch peak to gene association

@marianogabitto Can you provide insights?

Posterior sampling with Messenger Pyro guides -> huge memory use.

To add my "solution" to this which uses infer.Predictive and checks once that the variable is on a per cell basis and not a global parameter. It records all samples....

Improve --loci argument options

I recognized an additional problem on macOS (I am not sure whether it's OS specific), where it needs two spaces between the different loci (--loci A B G D works...

TypeError with model.train() using R reticulate

Can you share the full code. It looks like you have defined a custom seed like scvi.settings.seed = 2. This 2 is interpreted as a float in R. However, scvi-tools...

TypeError with model.train() using R reticulate

Closed due to inactivity.

scvi tools train with multi gpus

Low CPU load is usually not a problem if you reach high GPU usage. It just tells you, that it's less work and doesn't require 30 workers. The copy speed...

Mixed usage of `n_samples_overall` in `get_normalized_expression`

@PierreBoyeau Currently this means that these other function don't use importance weighting? I assume it would make sense to add it to these models as well?

Mixed usage of `n_samples_overall` in `get_normalized_expression`

I think we can improve the naming as well (but that's styling and I would stick to n_samples_overall for number of mc_samples and n_samples to number of samples for a...