Brief¶

Take a look at the Default parameters¶

Default parameters

Easy usage¶

By default, the database construction, model training, and model application are controled by the Default parameters .

For example:

# Build database
from agat.data import BuildDatabase
if __name__ == '__main__':
    database = BuildDatabase()
    database.build()

# Train AGAT model
from agat.model import Fit
f = Fit()
f.fit()

# Application (high-throughput prediction)
from agat.app.cata import HtAds
ha = HtAds()
ha.run(formula='NiCoFePdPt')

As you can see, you only need to provide very few input arguments to instantiate the imported objects.

Customized usage¶

There are generally two ways to customize your AGAT training and deployment. Taking the training as an example:

Customized keyword arguements¶

from agat.model import Fit
f = Fit(verbose=2, gat_node_dim_list=[6, 10, 10], b=10.0)
f.fit()

This will replace the settings for verbose, gat_node_dim_list, b. Other parameters are still controled by the Default parameters .

Customized input dict¶

import os
import torch.nn as nn
from agat.model import Fit
train_config = {
    'verbose': 1, # `0`: no train and validation output; `1`: Validation and test output; `2`: train, validation, and test output.
    'dataset_path': os.path.join('dataset', 'all_graphs.bin'),
    'model_save_dir': 'agat_model',
    'epochs': 1000,
    'output_files': 'out_file',
    'device': 'cuda:0',
    'validation_size': 0.15,
    'test_size': 0.15,
    'early_stop': True,
    'stop_patience': 300,
    'gat_node_dim_list': [6, 100, 100, 100],
    'head_list': ['mul', 'div', 'free'],
    'energy_readout_node_list': [300, 300, 100, 50, 30, 10, 3, 1],
    'force_readout_node_list': [300, 300, 100, 50, 30, 10, 3],
    'stress_readout_node_list': [300, 300, 6],
    'bias': True,
    'negative_slope': 0.2,
    'criterion': nn.MSELoss(),
    'a': 1.0,
    'b': 1.0,
    'c': 0.0,
    'optimizer': 'adam', # Fix to sgd.
    'learning_rate': 0.0001,
    'weight_decay': 0.0, # weight decay (L2 penalty)
    'batch_size': 64,
    'val_batch_size': 400,
    'transfer_learning': False,
    'trainable_layers': -4,
    'mask_fixed': False,
    'tail_readout_no_act': [3,3,3],
    'adsorbate_coeff': 20.0 # indentify and specify the importance of adsorbate atoms with respective to surface atoms. zero for equal importance.
    }
f = Fit(**train_config)
f.fit()

The input train_config has higher priority than the Default parameters .