Improvements self-issue

[anuradhawick]

Computations

• Add checkpoint between counting 15mers and generating coverage histogram profiles.
• Combine the steps and put a check for the computed 15mer counts. Load or compute and save. Avoid saving+writing cost
• Use c++ valarray for assignment step (may be compiler will use SSE)
• Add compiler optimization flags -O3 (make last assignment step faster)
• Add evaluation step to the final assigned bins (there is a difference between large enough bins and classifications.txt file) Fix it!

Binner

• Organize different embeddings into different classes
• Provide composition-only and coverage-only options
• Adjust re-sampling to suit different embedding strategies.
• Maybe steal ideas from LRBinner as a pre-step for embedding. VAE+UMAP/SONG works extremely well!
• High-dimensional noise filtering using LRBinner algorithm (should help with noise in ONT reads)

[anuradhawick]

• [x] Add checkpoint between counting 15mers and generating coverage histogram profiles.
• [x] Use c++ valarray for assignment step (may be compiler will use SSE)\
• [X] Add compiler optimization flags -O3 (make last assignment step faster)