feat: Integration of torch models in main #34
Closed
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
This commit adds the following implementations to the models_torch subpackage: - Added __init__.py for the subpackage - Implemented GraphAttentionNetwork in graph_attention_network.py - Implemented GraphConvolutionalNetwork in graph_convolutional_network.py - Implemented GraphSAGENetwork in graph_sage_network.py - Implemented GraphTransformerNetwork in graph_transformer_network.py These implementations provide configurable nn architectural support for training graph-based models using PyTorch.
Create a new directory named jaqpotpy_torch/ to organize all torch-related code. We'll decide later whether to keep this code here or move it to an entirely new standalone torch-specific package.
This commit initializes the featurizers_torch subpackage, adding the following implementations: - Added __init__.py for the subpackage - Implemented SmilesGraphFeaturizer in smiles_graph_featurizer.py. The SmilesGraphFeaturizer class is designed to create custom graph featurizations from SMILES strings. It offers highly configurable options, allowing users to choose from a wide range of both atom and bond characteristics to be included.
This commit initializes the datasets_torch subpackage, adding the following implementations: - Added __init__.py for the subpackage - Implemented SmilesGraphDataset in smiles_graph_dataset.py. The SmilesGraphDataset class is designed to create a custom torch Dataset for graph-featurized SMILES. Its __getitem__ method is overridden to return a torch_geometric Data object enacpsulating the following information: - Node attributes (x) - Edge indices (edge_index) - Edge attributes (edge_attr) - Target labels (y) - The original SMILES representation (smiles) SmilesGraphDataset enables straightforward integration into torch-based ML pipelines, facilitating the development of graph-based predictive models.
This commit removes the _torch suffix directory names within the jaqpotpy_torch module: - Renamed datasets_torch directory to datasets - Renamed featurizers_torch directory to featurizers - Renamed models_torch directory to models
This commit initializes the trainers subpackage, adding the following implementations: - Added __init__.py for the subpackage - Implemented an initial version of TorchModelTrainer abstract class in torch_model_trainer.py.
This commit initializes the trainers subpackage, adding the following implementations: - Added BinaryGraphModelTrainer, RegressionGraphModelTrainer in __init__.py - Extended TorchModelTrainer class with additional attributes.
This commit adds the following implementations to the trainers subpackage: - BinaryGraphModelTrainer subclass - RegressionGraphModelTrainer subclass
This commit adds the SmilesGraphDatasetWithExternal class implementation to the datasets subpackage. This class inherits from SmilesGraphDataset, and adds the functionality of providing an external feature vector along with the smiles representation.
This commit adds the implementation of the Featurizer abstract class. Also the abstract method featurize() is defined.
This commit adds the FullyConnectedNetwork class implementation to the models.
This commit adds the following implementations to the models subpackage: - GraphAttentionNetworkWithExternal - GraphConvolutionalNetworkWithExternal - GraphSAGENetworkWithExternal - GraphTransformerNetworkWithExternal In these models the corresponding graph neural network is employed to produce global level representations from smiles. Then these are concatenated with the external feature vectors and the concatenated vector is passed through a fully connected network to produce the final output.
- Fixed a circular import error of the FullyConnectedNetwork class - Added super().__init__() to all the models supporting external features
This commit adds the implementation of the deploy_model for both RegressionGraphModelTrainer and BinaryGraphModelTrainer. deploy_model() is an abstract method of the TorchModelTrainer base class, and must be implemented in every class that inherits from TorchModelTrainer, to support model deployment on Jaqpot.
In this commit we: - Implement the deployment logic for models and trainers that use external features - Set deploy_model function to be on the TorchModelTrainer class - Define the abstract method prepare_for_deployment() with a dynamic set of arguments per trainer subclass which transforms the data into the appropriate JSON - Add 'SMILES' in a protected namespace so that external features can't be named like this - Fix bugs regarding model input arguments
…com/ntua-unit-of-control-and-informatics/jaqpotpy into feat/JAQPOT-62/torch-graph-training
This commit provides: - Ready for deployment torch models are implemented - Everything up to date with the current structure of the API
This commit implements: - TabularDataset class inheriting from torch.utils.data.Dataset - BinaryFCModelTrainer & RegressionFCModelTrainer - The required changed in the BinaryModelTrainer and RegressionModelTrainer abstract classes to support data from torch.utils.data.DataLoader as well
This commit: - Adds example code blocks in SmilesGraphFeaturizer - Sets '.' instead of '_' for the separator when showing atom/bond characteristics vector labels - Add __call__ to to call abstract featurize() method in abstract class Featurizer
This commit: - Adds/refactors docs for models - Modifies author section
…com/ntua-unit-of-control-and-informatics/jaqpotpy into feat/JAQPOT-62/torch-graph-training
jpitoskas
changed the title
|
Closing this as it has been integrated into the main branch as part of JAQPOT-254 and is partly done on #47 |
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.
Incorporate Torch Model training into jaqpotpy
The
jaqpotpy_torchsupports the following types of modelling tasks:Notes for future implementation
Featurizers
Featurizers that inherit from our base featurizer class
Featurizermust implement thefeaturizemethod.Datasets
Datasets must inherit from Torch Dataset and implement/override the
__len__and__getitem__methods.Trainers
Trainers that inherit from our base class
TorchModelTrainermust implementget_model_type,train,evaluate,prepare_for_deployment.Models
Models must inherit form
nn.Moduleand implement__init__andforward.