Pytorch distributed training execution

PyTorch Distributed Training Execution

License: restricted — no clear open-source license detected for the underlying tool; verify licensing before commercial use or redistribution. <!-- asb-license-banner -->

Summary

Execute transfer learning workflows on graph neural network models using PyTorch and PyG with command-line parameter passing to enable fine-tuning on domain-specific molecular datasets. This skill bridges pre-trained model checkpoints and transfer learning datasets through scripted execution with structured dataset arguments.

When to use

When you have a pre-trained GNN model checkpoint and need to apply transfer learning to a new chromatography or molecular property prediction dataset (e.g., Eawag_XBridgeC18_364.xlsx) by fine-tuning the model weights on domain-specific examples without retraining from scratch.

When NOT to use

• Input dataset is already a learned feature embedding or pre-computed molecular fingerprint table (use direct model inference instead)
• No pre-trained model checkpoint is available (train from scratch using train_SMRT.py rather than transfer learning)
• Dataset is in a format other than .xlsx and conversion is not feasible

Inputs

• Pre-trained model checkpoint (PyTorch .pt or .pth file)
• Transfer learning dataset in Excel format (.xlsx)
• train_transfer_FE.py script
• Command-line arguments (--DataSet flag with dataset filename)

Outputs

• Fine-tuned model checkpoint
• Training logs with metrics (loss, validation scores)
• Model weights optimized for transfer learning dataset

How to apply

Load the pre-trained model checkpoint using PyTorch's model serialization and the transfer learning dataset (in .xlsx format) via Pandas. Execute the designated transfer learning script (train_transfer_FE.py) with the target dataset name passed as a command-line argument via the --DataSet flag. Use PyTorch optimizers and PyG's graph operations to perform forward/backward passes on molecular graph representations. Monitor training metrics (loss, validation performance) via TorchMetrics during fine-tuning epochs. Upon convergence or completion, save the fine-tuned model checkpoint and training logs for downstream evaluation.

Related tools

• PyTorch (Deep learning framework for model training, checkpoint loading, and gradient-based optimization)
• PyG (PyTorch Geometric) (Graph neural network operations for molecular graph representation and message passing)
• Pandas (Loading and parsing transfer learning datasets from Excel (.xlsx) files)
• TorchMetrics (Monitoring and logging training metrics during fine-tuning)
• torch-scatter, torch-sparse, torch-cluster (Graph operations and sparse tensor handling for PyG computations)
• RDKit (Molecular structure parsing and feature extraction for input preprocessing)

Examples

python train_transfer_FE.py --DataSet Eawag_XBridgeC18_364.xlsx

Evaluation signals

• Training loss decreases monotonically or shows expected convergence pattern across epochs, indicating model is learning dataset-specific features
• Validation metrics (e.g., RMSE, correlation) on held-out examples improve relative to the pre-trained baseline
• Fine-tuned model checkpoint is successfully saved and can be loaded without serialization errors
• Training logs from TorchMetrics contain expected fields (epoch, loss, validation score) with numeric values in plausible ranges
• Model inference on the transfer learning dataset produces predictions (e.g., retention time, solubility) compatible with input molecular properties

Limitations

• Transfer learning script assumes pre-trained model architecture matches the target dataset's molecular graph structure; mismatch will cause shape errors
• Dataset must be in .xlsx format; other formats (CSV, parquet) require conversion step not automated by this workflow
• No built-in hyperparameter tuning (learning rate, batch size, epochs) — these must be set manually in the train_transfer_FE.py script
• Training is executed sequentially on a single device; distributed multi-GPU execution is not documented in provided README

Evidence

• [intro] Transfer learning is executed by running train_transfer_FE.py with --DataSet flag: "python train_transfer_FE.py --DataSet Eawag_XBridgeC18_364.xlsx"
• [other] Pre-trained model and dataset are loaded, then script is executed with PyTorch and PyG for graph operations: "Load the pre-trained model checkpoint and transfer-learning dataset (Eawag_XBridgeC18_364.xlsx) using Python and Pandas. Execute train_transfer_FE.py with the --DataSet parameter set to"
• [other] Training metrics are monitored via TorchMetrics and results are saved: "Monitor training metrics via TorchMetrics during fine-tuning. Save the fine-tuned model checkpoint and training logs."
• [intro] Transfer learning is performed on thirteen datasets: "run the transfer learning on thirteen transfer learning data sets"
• [readme] Required dependencies include PyTorch, PyG, and graph-specific tensor libraries: "Pytorch, PyG, torch-scatter, torch-sparse, torch-cluster, torch_geometric"