Fixed n_jobs errors for umap-learn

[pteridin]

The warning UserWarning: n_jobs value -1 overridden to 1 by setting random_state. Use no seed for parallelism. occurs even when random_state is not set (= NA).

reticulate::use_condaenv("r-reticulate")
reticulate::py_install("umap-learn")

set.seed(123)
ds_train <- matrix(nrow = 1000,
                   ncol = 50,
                   runif(1000 * 50))

umap_model <- umap::umap(ds_train,
                         method = "umap-learn",
                         n_neighbors = 2,
                         n_components = 2,
                         metric = "cosine",
                         min_dist = .1,
                         spread = 1,
                         n_jobs = 12L,
                         
                         init = "random",
                         random_state = NA,
                         transform_state = NA,
                         verbose = TRUE)

This request fixes the error messages and allows parallel jobs.

[tkonopka]

Thank you @pteridin! This will be a useful feature.

I'll test this out and take a closer look at the weekend. In meantime, could I ask you please to bump the version number more, to 0.2.11.0? (I used the last digit only for attempts at cran submits)

[pteridin]

No problem.

Though, I must say that in my small little experiments: n_jobs>1 does not really bump the overall performance as much as I would have expected.

[tkonopka]

Code looks good, test suite passes, warning goes away for umap-learn (v0.5.6), and seeding is still active otherwise.

Great work, thank you for the contribution!