Leiden clustering for the cosine metric fails on the test data

[hoferand]

When enabling the cosine metric for Leiden clustering the analysis fails:

Traceback (most recent call last):
  File "XXX/unsupervised_analysis/.snakemake/scripts/tmptesnmdii.leiden_cluster.py", line 155, in <module>
    partition = la.find_partition(
  File "XXX/unsupervised_analysis/.snakemake/conda/58917ab1dd7b5a7c636be5faaa7d439f_/lib/python3.9/site-packages/leidenalg/functions.py", line 81, in find_partition
    partition = partition_type(graph,
  File "XXX/unsupervised_analysis/.snakemake/conda/58917ab1dd7b5a7c636be5faaa7d439f_/lib/python3.9/site-packages/leidenalg/VertexPartition.py", line 840, in __init__
    self._partition = _c_leiden._new_RBConfigurationVertexPartition(pygraph_t,
BaseException: Could not construct partition: Cannot accept negative weights.
RuleException:
CalledProcessError in file XXX/unsupervised_analysis/workflow/rules/clustering.smk, line 32:
Command 'source XXX/miniforge3/bin/activate 'XXX/unsupervised_analysis/.snakemake/conda/58917ab1dd7b5a7c636be5faaa7d439f_'; set -euo pipefail;  python XXX/unsupervised_analysis/.snakemake/scripts/tmptesnmdii.leiden_cluster.py' returned non-zero exit status 1.
Error in rule leiden_cluster:
    jobid: 40
    input: test/data/digits_data.csv, test/results/unsupervised_analysis/digits/UMAP/UMAP_cosine_50_graph.pickle
    output: test/results/unsupervised_analysis/digits/Leiden/Leiden_cosine_5_RBConfigurationVertexPartition_1_clustering.csv
    log: logs/rules/leiden_digits_cosine_5_RBConfigurationVertexPartition_1_clustering.log (check log file(s) for error details)
    conda-env: XXX/unsupervised_analysis/.snakemake/conda/58917ab1dd7b5a7c636be5faaa7d439f_

Software version: main@a86e42b

Config (all omitted keys are set to default):

umap:
  metrics: ["euclidean", "correlation", "cosine"]
  n_neighbors: [5, 15, 50]
  min_dist: [0.1]
  n_components: [2, 3]
  densmap: 1
  connectivity: 1
  diagnostics: 1

leiden:
  metrics: ["euclidean", "correlation", "cosine"]
  n_neighbors: [5, 15, 50]
  partition_types:
    ["RBConfigurationVertexPartition", "ModularityVertexPartition"]
  resolutions: [0.5, 1, 1.5, 2]
  n_iterations: 2

[sreichl]

maybe related: lmcinnes/umap#669
-> nope output_metric relates to the embedding space not the distance metric for KNN graph generation