Fix DR baselines

[scottgigante-immunai]

This PR fixes various problems with the dimensionality reduction baselines.

• density_preservation was giving a correlation of 0.5 when comparing the true data to itself
• spectral distances should be optimized by spectral embeddings
• spectral distances should weight the eigenvectors by their eigenvalues
• Removed some unnecessary baselines
• Added a diffusion maps method (incidental, through spectral distance fix) and a zebrafish dataset (as all existing datasets were rather simple)

[github-actions]

[scottgigante-immunai]

Closes #803

[lazappi]

Code looks good to me. Do you mind quickly explaining the motivation for the spectral distances? I'm just not very familiar with diffusion maps and would be good to have some background in case it comes up later.

[codecov]

Codecov Report

Base: 95.61% // Head: 95.68% // Increases project coverage by +0.07% 🎉

Coverage data is based on head (366bda8) compared to base (2da81a9).
Patch coverage: 98.11% of modified lines in pull request are covered.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #816      +/-   ##
==========================================
+ Coverage   95.61%   95.68%   +0.07%     
==========================================
  Files         184      186       +2     
  Lines        4949     4984      +35     
  Branches      273      271       -2     
==========================================
+ Hits         4732     4769      +37     
  Misses        138      138              
+ Partials       79       77       -2     

Flag	Coverage Δ
unittests	95.68% <98.11%> (+0.07%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
...tasks/dimensionality_reduction/methods/baseline.py	89.65% <85.71%> (-2.46%)	⬇️
...sks/dimensionality_reduction/datasets/zebrafish.py	100.00% <100.00%> (ø)
.../dimensionality_reduction/methods/diffusion_map.py	100.00% <100.00%> (ø)
.../tasks/dimensionality_reduction/metrics/density.py	100.00% <100.00%> (+4.65%)	⬆️
...ionality_reduction/metrics/distance_correlation.py	100.00% <100.00%> (ø)
test/test_task_dimensionality_reduction.py	100.00% <100.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[scottgigante-immunai]

Sure! Tl;dr diffusion distances are graph distances that are robust to noise. Where a graph shortest path distance is sensitive to a single noisy point short-circuiting the path between two otherwise distant points on the manifold, diffusion distances use the probability of traversing the graph from one point to another via a random walk to integrate over all graph paths instead of simply the shortest one.

[mvinyard]

Looks great, thank you @scottgigante-immunai !