Code

Graphium 3.0#519

Code

Draft

DomInvivo wants to merge 189 commits intomaindatamol-io/graphium:mainfrom graphium_3.0datamol-io/graphium:graphium_3.0Copy head branch name to clipboard

+12,323-12,965

Commits

Commits on Apr 13, 2024

• Added and integrated C++ graphium_cpp library, a Python module implemented in C++ for featurization and preprocessing optimizations, along with a few other optimizations, significantly reducing mem…

  Show description for 5ffe261
  ndickson-nvidia

  committed
  5ffe261

Commits on Apr 17, 2024

• Small changes to support not needing label data during data loading

  ndickson-nvidia

  committed
  8286383

• Removed FakeDataset, FakeDataModule, and SingleTaskDataset. SingleTaskDataset is still used in test_dataset.py, but won't after it's changed in a later commit.

  ndickson-nvidia

  committed
  dca9b2b

• Removed option to featurize using Python, (but didn't delete everything from the Python featurization yet), removed option to featurize to GraphData class instead of PyG Data class, added deprecati…

  Show description for 8304210
  ndickson-nvidia

  committed
  8304210

Commits on Apr 18, 2024

• Removed newly deprecated options from yaml files

  ndickson-nvidia

  committed
  4ee35d4

• Added support for limiting the number of threads used by prepare_and_save_data

  ndickson-nvidia

  committed
  cf23e37

• Fixed compiler warning about signed vs. unsigned comparison

  ndickson-nvidia

  committed
  5db0e2a

• Fixed Python syntax issues

  ndickson-nvidia

  committed
  c75a452

• Changed asymmetric inverse normalization type to be implemented using symmetric diagonalization, avoiding the need to handle complex eigenvectors and eigenvalues

  ndickson-nvidia

  committed
  4aa1f85

• Fixed compile errors

  ndickson-nvidia

  committed
  c53451a

Commits on Apr 19, 2024

• Some simplification in collate.py

  ndickson-nvidia

  committed
  268e245

• Deleting most of the Python featurization code

  ndickson-nvidia

  committed
  e032e8e

Commits on Apr 23, 2024

• Implemented conformer generation in get_conformer_features, trying to match behaviour from get_simple_mol_conformer Python code, but adding Hs, as recommended for conformer generation.

  ndickson-nvidia

  committed
  bdefe89

• Deleted deprecated properties.py

  ndickson-nvidia

  committed
  5298444

Commits on Apr 25, 2024

• Handle case of no label data in prepare_and_save_data. Also added concatenate_strings function, though it's not used yet.

  ndickson-nvidia

  committed
  c38aa06

• Changed prepare_data to support having no label data

  ndickson-nvidia

  committed
  86abf21

Commits on Apr 26, 2024

• Removed ipu metrics, since not compatible with latest torchmetrics

  DomInvivo

  committed
  bd59098

• Updated MetricWrapper to work with update and compute, compatible with torchmetrics

  DomInvivo

  committed
  734ba55

• Changed requirements for torchmetrics

  DomInvivo

  committed
  b6c578f

• fixed the loss by adding MetricToTorchMetrics, and added a few comments on the Predictor for Todos

  DomInvivo

  committed
  4f6e816

Commits on May 2, 2024

• Updated license passed to setup call in setup.py

  ndickson-nvidia

  committed
  80276da

Commits on May 3, 2024

• Major updates to predictor_summaries.py

  DomInvivo

  committed
  7933ae5

• Improved the predictor summaries. Added GradientNormMetric

  DomInvivo

  committed
  5849927

Commits on May 6, 2024

• Changes to get test_dataset.py and test_multitask_datamodule.py passing

  ndickson-nvidia

  committed
  9492e62

• Removed load_type option from test_training.py, because it's no longer used

  ndickson-nvidia

  committed
  d94097c

Commits on May 14, 2024

• Updated comment in setup.py about how to build graphium_cpp package

  ndickson-nvidia

  committed
  11e6935

• Rewrote test_featurizer.py. Fixed bug in mask_nans C++ function, and changed create_all_features to create all tensors even if there are nans, so that the number of atoms can still be determined fr…

  Show description for ff93c2d
  ndickson-nvidia

  committed
  ff93c2d

Commits on May 23, 2024

• Removed deprecation warnings and deprecated parameters from datamodule.py, keeping the notes in the function comments

  ndickson-nvidia

  committed
  a892068

• Recommended tweaks to extract_labels in multilevel_utils.py

  ndickson-nvidia

  committed
  38a5510

• Fixed "else if"->"elif"

  ndickson-nvidia

  committed
  f7771b3

Commits on May 24, 2024

• Rewrote test_pe_nodepair.py to use graphium_cpp

  ndickson-nvidia

  committed
  4256839

• Rewrote test_pe_rw.py to use graphium_cpp. Comment update in test_pe_nodepair.py

  ndickson-nvidia

  committed
  91c37a3

• Rewrote test_pe_spectral.py to use graphium_cpp

  ndickson-nvidia

  committed
  f347a0d

• Removed tests/test_positional_encodings.py, because it's a duplicate of test_pe_nodepair.py, apart from a couple comments

  ndickson-nvidia

  committed
  26b5531

Commits on May 28, 2024

• Fixed handling of disconnected components vs. single component for laplacian eigenvector/eigenvalue features, especially edge case with a single atom that was being treated as having multiple compo…

  Show description for 1ded38b
  ndickson-nvidia

  committed
  1ded38b

• Fixed compile warnings in one_hot.cpp

  ndickson-nvidia

  committed
  314d636

• Rewrote test_positional_encoders.py, though it's still failing the test_laplacian_eigvec_with_encoder case

  ndickson-nvidia

  committed
  e49b4da

Commits on Jun 4, 2024

• Removed commented out lines from setup.py

  ndickson-nvidia

  committed
  f001464

• Ran linting on Python files

  ndickson-nvidia

  committed
  2782fbc

Commits on Jun 5, 2024

• Hopefully explicitly installing graphium_cpp fixes the automated testing for now

  ndickson-nvidia

  committed
  77d27b5

• Test fix

  ndickson-nvidia

  committed
  cb1df19

• Another test fix

  ndickson-nvidia

  committed
  f3f6a0d

• Another test fix

  ndickson-nvidia

  committed
  c5c0085

• Make sure RDKit can find Boost headers

  ndickson-nvidia

  committed
  6dd827f

Commits on Jun 12, 2024

• Reimplemented test_pos_transfer_funcs.py to test all supported conversions

  ndickson-nvidia

  committed
  59c84a2

• Linting fixes

  ndickson-nvidia

  committed
  7bc8ade

• Fixed collections.abs.Callable to typing.Callable for type hint

  ndickson-nvidia

  committed
  6903243

Commits on Jun 13, 2024

• Improved the task summaries and started to fix the training logging.

  DomInvivo

  committed
  f355eed

Commits on Jun 17, 2024

• Removed file_opener and its test

  ndickson-nvidia

  committed
  9f38afb

Commits on Jul 9, 2024

• Fixed the issue with boolean masking, introduced by F._canonical_mask in TransformerEncoder

  DomInvivo

  committed
  5ab9ca9

• Fixed the float vs double issue in laplacian pos encoding

  DomInvivo

  committed
  9c7504f

• Added comment

  DomInvivo

  committed
  f8358f3

• Fixed the ipu tests by making sure that IPUStrategy is not imported if not using IPUs

  DomInvivo

  committed
  692decc

• Update test.yml to only test python 3.10

  DomInvivo

  authored
  8891e66

• Removed positional encodings from the docs

  DomInvivo

  committed
  c2d3c87

• Merge remote-tracking branch 'origin/dom_unittest' into dom_unittest

  DomInvivo

  committed
  d3d19d7

• Upgraded python versions in the tests

  DomInvivo

  committed
  0a1696f

• Removed reference to old files now in C++

  DomInvivo

  committed
  50265df

• Downgraded python version

  DomInvivo

  committed
  58fc2aa

• Fixed other docs broken references

  DomInvivo

  committed
  5852467

• Merge pull request #1 from ndickson-nvidia/dom_unittest

  Show description for ea9a775
  ndickson-nvidia

  authored
  ea9a775

• Merge pull request #510 from ndickson-nvidia/graphium_cpp

  Show description for 7f933b7
  DomInvivo

  authored
  7f933b7

Commits on Jul 10, 2024

• Fixed test_metrics. Moved lots of spaces.py imports to inner functions to avoid import loops

  DomInvivo

  committed
  4372ace

Commits on Jul 11, 2024

• duplicated some unit-test fixes from graphium_3.0 branch

  DomInvivo

  committed
  7b89998

• Fixed the loading of a previous dummy model using older metrics by removing the metrics from the model

  DomInvivo

  committed
  ab88952

• Minor documentation

  DomInvivo

  committed
  6c58733

• Removed the loss from predictor_summaries

  DomInvivo

  committed
  a9a8810

• Removed epochs from task summaries

  DomInvivo

  committed
  2185697

• Draft implementing the update/compute logic in the predictor.

  DomInvivo

  committed
  d37d818

• Fix the std metric. Still needs testing.

  DomInvivo

  committed
  b4524f9

• fixed all errors arising in test_finetuning.py

  DomInvivo

  committed
  5040c47

Commits on Jul 12, 2024

• Fixed the test_training.py unit test

  DomInvivo

  committed
  e761e08

• Standardized the test names

  DomInvivo

  committed
  5d60fbf

• Fixed some unit-tests that were broken by previous changes

  DomInvivo

  committed
  b59428a

• Added pytdc to the tests

  DomInvivo

  committed
  632d4dc

• Changed mamba install tdc to pip install, in the test.yml file

  DomInvivo

  committed
  0fa2d86

• Added '--no-deps' to TDC installation in test.yml

  DomInvivo

  committed
  2441f43

• Woops

  DomInvivo

  committed
  326b6e7

• Fixed issue with building docs

  DomInvivo

  committed
  641fa37

• Removed old file from breaking docs building

  DomInvivo

  committed
  2b85dce

• Changed to micromamba to install pytdc

  DomInvivo

  committed
  0c93a0f

• Added tests for the STDMetric and GradientNormMetric and fixed their outputs

  DomInvivo

  committed
  ec235fc

• Implemented test of MultiTaskSummaries. Only an error left for the mean and std of preds and targets

  DomInvivo

  committed
  38d03e1

Commits on Jul 13, 2024

• Fixed the preds and targets that were inverted in TaskSummary

  DomInvivo

  committed
  d6f62a4

• Tried to add grad_norm to the metrics, but won't work because it's not per task

  DomInvivo

  committed
  3673884

• Moved the gradient metric directly to the Predictor

  DomInvivo

  committed
  29598a2

• Removed file_opener and read_file

  DomInvivo

  committed
  6260fa1

• Fixed predictor grad_norm

  DomInvivo

  committed
  10a1017

• Merge branch 'graphium_3.0' into torchmetrics

  DomInvivo

  authored
  8aa0f2b

Commits on Jul 15, 2024

• Fixed the progress bar logging to newest version. Fixed minor issues in the logging and imports.

  DomInvivo

  committed
  90c0ca4

• Merge remote-tracking branch 'origin/torchmetrics' into torchmetrics

  DomInvivo

  committed
  be99d94

• fixed some issue with older version of torchmetrics

  DomInvivo

  committed
  44b66b5

Commits on Jul 16, 2024

• Fixed reversed preds/targets. Fixed random sampling to take in the DF idx. Few other minor changes

  DomInvivo

  committed
  5c421a6

• fixed missing metrics computation on on_train_batch_end

  DomInvivo

  committed
  f15cd9a

• Added toymix training to the unit-tests. Also useful to run in debug mode.

  DomInvivo

  committed
  2142313

• Adding _global/ to some metrics logging into wandb

  DomInvivo

  committed
  99e0cd6

• Added better handling of metrics failure with logger.warn

  DomInvivo

  committed
  045ea53

• Fixed metric issues on gpu by casting to the right device prior to .update

  DomInvivo

  committed
  d8ba606

Commits on Jul 17, 2024

• Added losses to the metrics, such that they are computed on val and test too

  DomInvivo

  committed
  1bf2734

• Restricting the numpy version due to issues with wandb

  DomInvivo

  committed
  68b9361

• detaching preds

  DomInvivo

  committed
  911dfe9

• Removed cuda version restriction

  DomInvivo

  committed
  d34ac60

• Removed unnecessary detach, that broke the loss

  DomInvivo

  committed
  b1f2e86

Commits on Jul 30, 2024

• Updating dep versions for bh2 install

  Andrewq11

  committed
  62b385a

Commits on Jul 31, 2024

• Merge pull request #522 from datamol-io/package/bh2-install

  Show description for 41490a0
  DomInvivo

  authored
  41490a0

Commits on Aug 8, 2024

• Fix lightning backend issue; add predict_step for inference

  WenkelF

  committed
  7f9112a

Commits on Aug 9, 2024

• Fixing device issue in metrics calculation

  WenkelF

  committed
  47b7d1c

Commits on Aug 15, 2024

• Minor gitignore

  DomInvivo

  committed
  9dbd021

Commits on Aug 16, 2024

• Fixed the error due to time metrics on CPU No backend type associated with device type cpu

  DomInvivo

  committed
  5a77cbe

• Added val epoch time

  DomInvivo

  committed
  7fba29d

Commits on Aug 17, 2024

• Added logic to avoid crashing when resetting unused metrics

  DomInvivo

  committed
  b59dc36

Commits on Aug 19, 2024

• Added MetricWrapper.device

  DomInvivo

  committed
  da3e3a1

• Disable caching model checkpoint through WandbLogger

  

  sft-managed

  committed
  8bf0d41

• Disabled caching model checkpoint through WandbLogger

  AnujaSomthankar

  committed
  1ec4969

• Drafting unit test for node ordering

  WenkelF

  committed
  9ba5a16

Commits on Aug 21, 2024

• Improved the testing of the metrics reset, update, compute

  DomInvivo

  committed
  6f35ea9

• Reverted wrong change in `train_finetune_test.py

  DomInvivo

  committed
  d2f84f2

• Improved __len__ in MultitaskDataModule

  DomInvivo

  committed
  e9be441

Commits on Aug 22, 2024

• Added a new logic to allow saving all preds and targets more efficiently for auroc and avpr

  DomInvivo

  committed
  eaf9077

• Fixed the concatenation to work with and without DDP. Moved to CPU for faster + less memory requirement

  DomInvivo

  committed
  5432531

• Fixed the issue with memory leaks and devices.

  DomInvivo

  committed
  8c75d77

• Fixed the CPU syncing of MetricToConcatenatedTorchMetrics and GPU for other metrics

  DomInvivo

  committed
  5abd769

• Fixed the training metrics, and grouped all epoch-time and tput metrics

  DomInvivo

  committed
  fac3052

• Fixing unit tests

  WenkelF

  committed
  6603014

• Fixed epoch_time tracking (because train ends after val)

  DomInvivo

  committed
  d0ed816

Commits on Aug 23, 2024

• Using the torchmetrics.Metric.sync instead of torch_distributed

  DomInvivo

  committed
  9b7063f

Commits on Aug 29, 2024

• Fixed issue that NaNs are always removed with mean-per-label

  DomInvivo

  committed
  136b8b0

• Changed the name of logging variables

  DomInvivo

  committed
  2724b4c

• Removed some IPU logic

  DomInvivo

  committed
  141f48b

• Fixed the syncing of MetricToConcatenatedTorchMetrics

  DomInvivo

  committed
  62f2224

• Fixed classification metric calculation when multitask_handling=flatten

  AnujaSomthankar

  committed
  2b58fed

Commits on Sep 5, 2024

• Partial fix of node label ordering

  WenkelF

  committed
  c23dc02

Commits on Sep 7, 2024

• Fixed all unit-test, except those for IPU

  DomInvivo

  committed
  2fb7f4b

• First pass at removing IPU

  DomInvivo

  committed
  607e71b

• More removal of ipu

  DomInvivo

  committed
  49e9984

• More removal of ipus

  DomInvivo

  committed
  d8786e9

• More removal of ipus

  DomInvivo

  committed
  6ed6bb8

• Remove packing

  DomInvivo

  committed
  495f3f6

• Fixing most unit-tests

  DomInvivo

  committed
  90c7af2

• Updated env file

  DomInvivo

  committed
  29229ff

• Fixed the dummy model, toymix run, and most unit-tests

  DomInvivo

  committed
  019be26

• Minor

  DomInvivo

  committed
  21a63a8

• Fixed all remaining unit-tests - mostly the attention layers

  DomInvivo

  committed
  4bbc1f9

• minor changes to env

  DomInvivo

  committed
  be17a1f

• Added comments to env file

  DomInvivo

  committed
  318694f

Commits on Sep 10, 2024

• Merge branch 'graphium_3.0' into torchmetrics

  AnujaSomthankar

  authored
  ce4f94d

• Merge pull request #517 from datamol-io/torchmetrics

  Show description for f723632
  AnujaSomthankar

  authored
  f723632

Commits on Sep 11, 2024

• Merge branch 'torchmetrics' into remove_ipu

  WenkelF

  authored
  c32be78

• Forcing gcc_linux_64 in the env file

  DomInvivo

  committed
  5097e2a

Commits on Sep 16, 2024

• Update test.yml

  DomInvivo

  authored
  29044f5

• Update test.yml

  DomInvivo

  authored
  1771b78

• Update test.yml

  DomInvivo

  authored
  f87ee26

Commits on Sep 19, 2024

• Changed persistent_workers=True to False

  AnujaSomthankar

  committed
  24412f6

Commits on Sep 23, 2024

• Reorder atoms in node-level and nodepair-level label data, when the same molecule appears in multiple tasks with different atom orders, to be consistent with the first task's atom order

  ndickson-nvidia

  committed
  1d7bfeb

• Merging of equivalent molecules is now optional, but still defaults to true.

  ndickson-nvidia

  committed
  b7d9fe7

• Fixed bug with recent change in smiles_to_brief_data

  ndickson-nvidia

  committed
  5932fd8

• Fix graphium_cpp.prepare_and_save_data call in test_dataset.py to include merge_equivalent_mols parameter

  ndickson-nvidia

  committed
  020b08a

• In MultitaskFromSmilesDataModule.get_data_hash, include options used during pre-processing

  ndickson-nvidia

  committed
  6cd5e26

• Linter fixes in python files already modified in this branch

  ndickson-nvidia

  committed
  d1cad44

• Split prepare_and_save_data into get_task_data, get_indices_and_strings, compute_mol_keys, compute_stats, save_non_label_data, and save_label_data, to make the code more manageable.

  ndickson-nvidia

  committed
  508abbd

• Added support for reordering edge label data if there are multiple tasks with the same molecule and one of them has edge-level label data. This should order bonds consistent with the first copy of …

  Show description for 044cd47
  ndickson-nvidia

  committed
  044cd47

• Changed parse_mol in graphium_cpp.cpp to order based only on explicit atom classes in the SMILES strings, and only if they represent a complete atom order, to fix the accidental removal of support …

  Show description for 50adb1b
  ndickson-nvidia

  committed
  50adb1b

• The datasets use 0-based indexing for explicit ordering via atom classes in SMILES strings, not 1-based

  ndickson-nvidia

  committed
  c870af4

• Started adding doxygen comments to the C++ code. Also changed compute_laplacian_eigendecomp to accept const int32_t* instead of const std::vector<int32_t>*

  ndickson-nvidia

  committed
  90f2403

• Adding unit test for node ordering

  

  WenkelF

  authored and

  ndickson-nvidia

  committed
  bafdfe8

• Added doxygen comments for functions and enums related to one-hot feature encoding

  ndickson-nvidia

  committed
  7fd40e7

• Added more doxygen comments

  ndickson-nvidia

  committed
  06b12b2

• Added and updated more comments

  ndickson-nvidia

  committed
  5d798a5

• Added comments to each function in features.cpp

  ndickson-nvidia

  committed
  7261279

• Investigating failing unit tests

  WenkelF

  committed
  b82f582

• Added more comments to labels.cpp

  ndickson-nvidia

  committed
  4a152b2

• Merge branch 'atom_order' of ssh://github.com/ndickson-nvidia/graphium into atom_order

  ndickson-nvidia

  committed
  618bbb1

• Build fix in features.cpp

  ndickson-nvidia

  committed
  92ab751

• Skipping test_training.py for now

  WenkelF

  committed
  f123565

• Merge pull request #521 from ndickson-nvidia/atom_order

  Show description for e887176
  DomInvivo

  authored
  e887176

Commits on Sep 27, 2024

• Updated documentation

  AnujaSomthankar

  committed
  1c4aa3b

• Updated documentation

  AnujaSomthankar

  committed
  75df01a

Commits on Oct 31, 2024

• Wrapping up finetuning updates

  WenkelF

  committed
  8436111

Commits on Nov 1, 2024

• Update readme

  WenkelF

  committed
  9cfabb7

• Fixed finetuning unit test

  WenkelF

  committed
  bf504a8

• Fixing docs

  WenkelF

  committed
  42c9dbc

Commits on Nov 4, 2024

• Merge pull request #530 from datamol-io/remove_ipu

  Show description for cbf9dc7
  WenkelF

  authored
  cbf9dc7

• Reducing size of example finetuning dataset

  WenkelF

  committed
  807913f

Commits on Nov 5, 2024

• Cleaning up configs and naming conventions

  WenkelF

  committed
  697e3d1

• Minor changes and documentation

  WenkelF

  committed
  d9aa407

• Minor change

  WenkelF

  committed
  7759266

• Merge pull request #529 from datamol-io/upgrade-finetuning

  Show description for 7cfa89a
  WenkelF

  authored
  7cfa89a

Commits on Nov 14, 2024

• Added C++ file description comments

  ndickson-nvidia

  committed
  8a65d84

Commits on Nov 16, 2024

• Merge pull request #532 from ndickson-nvidia/cpp_file_descriptions

  Show description for 5ffc2f6
  DomInvivo

  authored
  5ffc2f6