Model Prediction Operation

After completing model training, you can use the trained model to predict Hamiltonians for new structures.

Preparation for Prediction

Construct graph_data.npz for structures to be predicted:
- Convert structures to the corresponding DFT software format
- Use post-processing tools (such as openmx_postprocess) to generate overlap.scfout
- Configure graph_data_gen.yaml, setting scfout_file_name to 'overlap.scfout'
- Run graph_data_gen to generate graph_data.npz
Edit configuration file: Modify the following parameters in config.yaml:
```
setup:
  checkpoint_path: /path/to/trained/model.ckpt  # Path to trained model
  num_gpus: null  # null or 0: CPU; positive integer: GPU count
  stage: test  # Set to test for prediction mode
```
accelerator supports null, 'cpu', 'gpu', and 'ddp'; num_gpus remains the device-count or device-ID setting. For training resumption, set resume: true and provide a non-empty checkpoint_path; HamGNN calls trainer.fit(model, datamodule, ckpt_path=checkpoint_path). This restores optimizer, scheduler, epoch, and global-step state, whereas load_from_checkpoint starts a new task from weights. precision: 32 uses standard precision and 64 uses double precision.
Set environment variables: For CPU execution, you can accelerate with multi-threading:
```
export OMP_NUM_THREADS=64  # Set number of threads
```

Execute Prediction

HamGNN --config config.yaml > predict.log 2>&1

For large systems or batch predictions, it’s recommended to submit tasks using a job scheduling system:

#!/bin/bash
#SBATCH --partition=compute
#SBATCH --job-name=HamGNN_predict
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
#SBATCH --cpus-per-task=64
#SBATCH --exclusive

export OMP_NUM_THREADS=64
source /path/to/your/miniconda3/bin/activate your_env_name

HamGNN --config ./config.yaml > predict.log 2>&1

Output Results

After prediction completion, the following files will be generated in the directory specified by train_dir:

prediction_hamiltonian.npy: Predicted Hamiltonian matrix in NumPy array format
target_hamiltonian.npy: If the input graph_data.npz contains the actual Hamiltonian matrix, this will be the true value from the input