Releases: seqcode/TPCAV
Release list
0.10.3
Minor bug fix, improving log messaging & html report
- Fix the index overflow bug when total elements of activation matrix reaches 2^31
- Improve log messaging and parameter names
- Add fscore threshold to motif ranking table to avoid poorly classified motif concept confuses the user
Full Changelog: 0.10.2...0.10.3
0.10.2
Optimize saving and loading of CavTrainer, avoid huge memory consumption when multiple instances are saved and loaded
Full Changelog: 0.10.1...0.10.2
0.10.1
New rust backend to optimize python functions
- thin_svd backed by rust faer to break the 32bit integer index limit of LAPACK
- sample_from_pwm in rust improves efficiency when sampling motif instances
What's Changed
New Contributors
Full Changelog: 0.9.11...0.10.1
0.9.11
Minor fix
- Now using pybedtools.BedTool to load bed file to avoid missing columns
- Fix the issue that the strand information of sequence bed concept is missing
Full Changelog: 0.9.10...0.9.11
0.9.10
Minor fix: put pca evaluation calculation on cpu to avoid OOM on GPU
Improvement: add download buttons for tables in html report
Full Changelog: 0.9.9...0.9.10
0.9.9
Add heuristic way to determine # samples per concept when it's not specified
Full Changelog: 0.9.8...0.9.9
0.9.8
Small bug fix, improving messaging, especially on reporting quality of PCA transformation
Full Changelog: 0.9.7...0.9.8
0.10.1a
Add multiple features by claude:
- Improve loading data from npy file to reduce memroy footprint by avoid array copy
- add new de_corr backend using a single linear layer with loss function on covariance matrix, this allows tpcav works at big models
- performance optimization at multiple places
Full Changelog: 0.9.6...0.10.1a
0.9.7
Improve memory usage when evaluating classifier
Full Changelog: 0.9.6...0.9.7
0.9.6
Fix a bug that storing attributed layer in tpcav model causes captum to register hook twice on the same module
Full Changelog: 0.9.5...0.9.6