James Melville

* In my experience Annoy does not perform well with high dimensional datasets: even with a large value of `n_trees` and `search_k` it can run for a long time and...

Hi Jianshu, > I want to revisit this topic. As I said before, after building the hnsw, k best neighbors can be extracted from hnsw graph bottom layer (layer 0)...

> I am not sure about the annoy time complexity because they refuse to provide one but for hnsw I am pretty sure because the > original paper provides details....

As usual, it's taken *years*, but RcppHNSW is now supported as an optional dependency (i.e. you must install it yourself). For now this functionality lives on the `hnsw` branch, but...

This has been merged and tested so should appear in the next CRAN release.

I believe the irlba/Matrix error will only occur if the C code (calling the `irlba` function with `fastpath = TRUE`) is executed. So `master` now contains an attempt to `tryCatch`...

I'm afraid I don't have access to a Windows 7 machine to test this. On Windows 11, I am unable to reproduce the problem. Does running: ```r utils::untar(model_file, list =...

It looks like the colon in the drive letter makes `tar` think it's a file on another machine. This changed on Windows 10 onwards according to `?utils::tar`. Does this work:...

I can attempt to fix the `load_uwot` and `save_uwot` functions by detecting if the user is running Windows 7. @mytarmail could you paste the output from `Sys.info()`?

Thank you @mytarmail. I have made a change to the save/load model code that attempts to detect Windows 7 and use the `--force-local` argument, so hopefully Windows 7 users won't...