Sebastian Bischoff
Sebastian Bischoff
Why does `transform` return different embeddings for the same data? ```julia model = UMAP_(transpose(data), 2; n_neighbors=15, min_dist=0.1) UMAP.transform(model, transpose(data[1:20,:])) ``` results in ```julia 2×20 Matrix{Float32}: -7.9815 8.59674 -7.18658 7.90244 …...
Is there a way to do the inverse transformation (like [https://umap-learn.readthedocs.io/en/latest/inverse_transform.html](https://umap-learn.readthedocs.io/en/latest/inverse_transform.html)) from coordinates in the UMAP projection back to the original space?
I would like to use the function `f(x)` like you describe it in the "10min to dynamo" tutorial [1] to get the velocity for unseen gene expression vectors `x`. Unfortunately,...
This PR currently just adds two tests, one were running `tl.cell_velocities` works and one were selecting for `transition_genes` fails. From trying to debug it, I would say that there is...
`zuko` as dependency requires at least python 3.8. Changed `pyproject.toml` and added sentence to `CONTRIBUTING.md`. sbi installed via `pip install sbi` also seems to need at least python 3.8 (in...
Add docker containers for easy usage. GPU containers not yet tested.
Add .gitignore for data files. Only tested for CUB as I couldn't download the other data yet.
Add requirement files to easily install the dependencies with pip and use python3 consistently. The code wasn't compatible with `pytorch=1.0.0`.
By checking if the system has CUDA installed, we can make the code compatible to systems with and without GPUs. This allows to develop on a system without GPU.
Varnish doesn't cache requests with the cookie because each one has another cookie