Could analyze_landscape auto-generate a Jupyter notebook like the cryoDRGN_filtering notebook created by cryodrgn_analyze?

[cbeck22]

Hi!

I've discovered some interesting non-linear trends in the volume PCA performed by analyze_landscape, and I want to try and analyze it further using non-linear dimensionality reduction techniques:

However, all of the volume PCA analysis for analyze_landscape occurs under the hood, and it's not possible to access the actual list of volumes directly to perform the PCA analysis as implemented in the analyze_volumes function in analyze_landscape.py:

vols = torch.stack(
    [
        ImageSource.from_file(
            f"{outdir}/kmeans1000/vol_{vol_start_index+i:03d}.mrc"
        ).images()[mask]
        for i in range(K)
    ]
)
vols[vols < 0] = 0

I was able to cobble together something that works by copying over the ImageSource, MRCFileSource and MRCHeader classes into a Jupyter notebook, but it would be convenient to be able to access the list of volumes similar to how we can access that z-values in the cryodrgn_filtering notebook so that I could more easily perform analyses like kernel PCA or UMAP. As an aside, here's the results of the UMAP analysis. I'm not really sure what it means yet, but I'm working on it...

Cheers,
cbeck

[cbeck22]

Update: I've realized that the notebook generated by cryodrgn_analyze_full is what I'm actually looking for. I would still appreciate the option to generate a similar notebook for cryodrgn_analyze.