LR scheduler for SGD in NeuroChem trainer #282
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Codecov Report
@@ Coverage Diff @@
## master #282 +/- ##
==========================================
+ Coverage 86.13% 86.15% +0.02%
==========================================
Files 15 15
Lines 1370 1372 +2
==========================================
+ Hits 1180 1182 +2
Misses 190 190
Continue to review full report at Codecov.
|
zasdfgbnm
added a commit
that referenced
this pull request
Sep 23, 2019
* Force coefficient is better to be 0.1 (#249) Being 1 does not work well * Change unidiomatic type calls to isinstance and remove deprecated h5py operator (#250) * Change unidiomatic type calls to isinstance and remove deprecated h5py operator * Add whitespace after , to make flake8 happy * Reformat the BuiltinModels and Builtins API to make them simpler and clearer (#252) * compute atomic self energies for a given data set prior to training (#255) * using least-squares to compute atomic self energies from the dataset * self atomic energy calculation in the example training file * import ignite and data separately (#256) * Support 0 molecule subsets when loading dataset (#257) * Support 0 molecule subsets when loading dataset * fix * fix * Simplify computation of strain (#260) * Update ase.py * Update ase.py * Scale by sqrt(natoms) instead of natoms (#259) * AdamW implementation minor fix (#261) * ASE dropped python2 support (#265) * Cleanup code for builtin models (#266) * Add "do not merge" blocker (#268) * Small doc fixes (#270) * calculate intercept when fitting, discard outliers from dataset (#263) * Replace BatchedANIDataset (#272) * Fix codefactor warnings (#273) * Fix pyanitools docstrings * Make offending ignite docstring raw * Make flake8 happy * Port to torch.bool data type where it should be (#278) * Port to torch.bool data type where it should be * more * more * fix snoop * hopefully fixed * fix ignite * remove snoop * flake8 * Add test for isolated atoms and large distances for 2 and 3 atoms (#274) * Adapt neurochem trainer to NC setup (#275) * Remove the first layer's weight decay parameters of each atomic network (#279) * LR scheduler for SGD (#282) * Update nnp_training to have identical setup to NC (#280) * Update nnp_training to have identical setup to NC * fix * add LR decay scheduler for both optimizers * fix * Update force training example (#283) * Update nnp_training to have identical setup to NC * fix * add LR decay scheduler for both optimizers * fix * update example files * fix * fix * enable discription in pypi page (#285) https://pypi.org/project/torchani/ * Support len in ChemicalSymbolsToInts, len of it returns number of supported elements (#286) * Support len in ChemicalSymbolsToInts * fix * flake8 * Discard outlier energy conformers (#287) * outlier removal process fixed * remove outlier energies if exist * small fix in methane, had a missing atom (#288) * Remove all large files and stop using git-lfs (#289) * Accelerate angular AEV computation and reduce memory cost (#290) * Accerate angular AEV computation and reduce memory cost * reduce number of elementwise product * Change download URL to dropbox (#291) * Update test_data.py (#292) * Update README.md installation (#294) * Remove unnecessary import (#296) * Remove unnecessary import * fix * New dataset API, cached dataset and shuffled dataset (#284) * Add split to new dataset API (#299) * split * clean * docs * docs * Update new.py * Fix dependency in setup.py: `torch-nightly` has been renamed to `torch` (#295) * Fix dependency in setup.py: `torch-nightly` has been renamed to `torch` `torch-nightly` has been renamed to `torch` * Update install_dependencies.sh * Update install_dependencies_python2.sh * Update deploy-docs.yml * Update docs.yml * Update tools.yml * Update setup.py * Update start.rst * Update runnable_submodules.yml * Update install_dependencies.sh * Update install_dependencies_python2.sh * Update start.rst * Update README.md * Update README.md * New Dataset API add other properties (#300) * cached * typo and comments * easy to read * change some names * fix unit test * empty line * fix * fix * add docs and add whether include_energies * docs * other properties for shuffled dataset * docs * dtype for benchmark * add properties to test * style * [JIT] Add TorchScript compatibility for AEVComputer (#303) * make aev,model compatible with jit * add type annotation to nn * flake8 fix * refactor AEVComputer * fix doc * an example with padding * use Optional type instead of padding * fix * fix * make pbc and cell keyword arguments in test_aev * fix * make pbc and cell keyword arguments in ase * fix * fix * fix dtype * fix * aev_computer dtype to double * change test files to have aev_computer with keyword argument * fix JIT types * add TestAEVJIT * fix LGTM alerts * fix TestAEVJIT * Update aev.py workaround for dtype in `torch.arange` * More arange bugs * Even more arange * fix LGTM alert * [JIT] Add TorchScript Compatibility for EnergyShifter (#306) * enable EnergyShifter scripting * fix * fix * Fix the dimension in self_energies for dataset containing only one elements (#302) * Update utils.py if I train molecule like Oxygen which contains only one element, the self.self_energies is a scalar and will make the operation self_energies = self.self_energies[species] crash. Besides, the species of Oxygen is [[3,3]] and is over the boundary of the energy list, so need to minus the minimum. * Update utils.py * Use a modified Sequential class with type annotation (#308) * Move flake8 check to github actions (#309) * Update and rename pythonpackage.yml to flake8.yml * Delete flake8.yml * Update flake8.yml * Move docs to github actions (#310) * Move docs to github actions * Delete docs.yml * Create install_dependencies.sh (#311) * Rename install_dependencies.sh to ci/install_dependencies.sh * +x * Move tools check to GitHub Actions (#314) * Move tools check to GitHub Actions * Delete tools.yml * Move submodules check to GitHub Actions (#313) * Create runnable_submodules.yml * Delete runnable_submodules.yml * Update runnable_submodules.yml
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.