Releases: ml-struct-bio/cryodrgn
Releases · ml-struct-bio/cryodrgn
Version 1.1.0
Updated default settings to larger model architecture, modified positional encoding, and accelerated training:
- Mixed precision training is now turned on by default (Use
--no-amp
to revert to single precision training) - Encoder/decoder architecture is now 1024x3 by default (Use
--enc-dim 256
and--dec-dim 256
to revert) - Gaussian Fourier featurization for faster training and higher resolution density maps (Use
--pe-type geom_lowf
to revert)
Version 1.0.0
Release for version 1.0.0
NEW: cryodrgn analyze_landscape
for automatic classification and energy landscape inference
NEW: Faster training and higher resolution model with Gaussian Fourier featurization (Use --pe-type gaussian
)
NEW: cryodrgn_utils <command>
-h for standalone utility scripts
NEW: cryodrgn_utils write_star
for converting cryoDRGN particle selections to .star
files
Add pytorch native mixed precision training and fix support for pytorch 1.9+
Version 0.3.4
Updates to auxiliary scripts
- FIX: Bug in
write_starfile.py
when provided particle stack is chunked (.txt file) - Support micrograph coordinates and additional column headers to
write_starfile.py
- New helper scripts:
analyze_convergence.py
(in beta testing) contributed by Barrett Powell (thanks!) andmake_train_test.py
for splitting up particle stacks for training
Version 0.3.3
Version 0.3.2
Minor updates to 0.3.1 software
- New: Additional Jupyter notebook for particles filtering,
cryoDRGN_filtering.ipynb
- New:
cryodrgn view_config
- Fix: Compatibility and deprecation fixes with pytorch 1.7 (#29), scipy (#39), and seaborn (3111efd)
- Updated: More functionality for converting to starfiles with
write_starfile.py
(#44) - Performance improvements to
cryodrgn eval_vol
and a minor fix for a corner case (#33) - Note: The logged KL divergence in stdout is no longer scaled by beta, however the default objective is identical
Version 0.3.1
Minor update to v0.3 software:
- New script
write_starfile.py
to save selected particles as a .star file for use in other tools - Improvements to
cryodrgn analyze
Version 0.3.0
- New: GPU parallelization with flag
--multigpu
- New: Mode for accelerated mixed precision training with flag
--amp
, available for NVIDIA tensor core GPUs - Interface update:
- Renamed encoder arguments
--qdim
and--qlayers
to--enc-dim
and--enc-layers
- Renamed decoder arguments
--pdim
and--players
to--dec-dim
and--dec-layers
- Renamed encoder arguments
- Argument default changes:
- Flipped the default for
--invert-data
to True by default - Flipped the default for
--window
to True by default
- Flipped the default for
- Updated training recommendations in below quick start guide
- Updates to cryodrgn analyze
- More visualizations
- Order kmeans volumes according to distances in latent space (previously random)
- More features for particle selection and filtering in the Jupiter notebook