计算数据预训练后再用实验数据微调，预测 logD7.4，代码：PCFE-GNNs，原文：Improved GNNs for Log D7.4 Prediction by Transferring Knowledge from Low-Fidelity Data，代码解析继续downstream_task。模型框架如下：

1.pretrain.py

if __name__ == "__main__":
    import argparse
    import os
    from configure import get_exp_configure

    parser = argparse.ArgumentParser(description='Regression model')
    parser.add_argument('-m', '--model', type=str,
                        choices=['GCN','GAT','Weave','AttentiveFP','MPNN', 'SchNet', 'MGCN'],
                        default='GCN',
                        help='Model to use')
    parser.add_argument('-mt', '--metric_name', type=str, default='R2',
                        help='Meter')
    parser.add_argument('-t', '--task', type=str, default='caco2',
                        help='data path')
    args = parser.parse_args().__dict__

    args['task'] = r'chembl_CXlogD_1711398.csv'

    args['model'] = 'AttentiveFP'
    args['weights_path'] = 'weights/AttentiveFP_chembl-CX_pretraining_weights.pth'

    args['smiles_columns'] = 'Smiles'
    args['task_names'] = ['CX LogD']

    args['exp'] = args['model']
    args.update(get_exp_configure(args['exp']))
    args['mode'] = 'higher'
    args['n_tasks'] = 1
    args['metric_name'] = 'R2'
    args['random_seed'] = 7

    os.environ["CUDA_VISIBLE_DEVICES"] = '0'
    args['device'] = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")

    if torch.cuda.is_available():
        print('GPU: ', os.environ["CUDA_VISIBLE_DEVICES"],
                  torch.cuda.get_device_properties(torch.cuda.current_device()))

    print(args['task'], '\t',args['model'], '\t', args['random_seed'])

    test_score, mse_score, mae_score = main(args)
    print('best R2:',test_score)
    print('best mse:',mse_score)
    print('best mae:',mae_score)