Dmitry Kobak

This is not currently possible in FIt-SNE and I have never seen this done in the literature. Sounds more like a research project.

Hi. The problems with Annoy should have all been fixed in the current version. Why couldn't you install 1.2.1? I did not really understand that part.

I see. Somebody should make a PR to push the latest 1.2.1 version to Seurat. @linqiaozhi Can you do it? Or maybe @rafalwoycicki you want to do it?

I was taking care to specify exaggeration correctly, so I think the only difference must have been due to adaptive gradient descent... Anyway, optional output sounds right. I noticed that...

OK. Currently the C++ code always saves the costs, but I'm not sure it's a good idea with gradient descent history: e.g. for 1mln points, the output would be 1mln...

@pavlin-policar I don't think there is a way to set up a callback system. My current thinking is that we should pass a boolean flag `save_intermediate_iterations` into the C++ code....

This means that the Seurat wrapper should be updated. Indeed, it does not support Fit-SNE 1.2: https://github.com/satijalab/seurat/blob/b51801bc4b1a66aed5456473c9fe0be884994c93/R/dimensional_reduction.R#L1820 Until this is fixed, you could install FIt-SNE 1.1 instead of 1.2.

I reopen this though, because it should be fixed in Seurat. Hopefully @linqiaozhi can take care of that.

No distance metric apart from the Euclidean is currently supported. So you can run it on binary data, but only using Euclidean distances. @linqiaozhi Annoy supports various metrics out of...

I don't know why this happens with `writeBin()`, but note that 33650 is way too high dimensionality to be useful and will likely only cause problems for `annoy` nearest-neighbours search....