inputs_2d=feat['inputs_2d'],
mask=feat['mask'],
affine=qa_single)
res_batch = attn_batch(inputs_1d=inputs_1d_batch,
inputs_2d=inputs_2d_batch,
mask=mask_batch,
affine=qa_batch)
print(check_recursive(res_single, res))
print(check_recursive(res_batch[0, ...], res))
for i in range(batch_size):
err = torch.sum(torch.abs(res_batch[i, ...] - res_single))
print(i, err.item())
assert err < 1e-2
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Train deep protein docking')
parser.add_argument('-debug_dir',
default='/home/lupoglaz/Projects/alphafold/Debug',
type=str)
args = parser.parse_args()
config = model_config('model_1')
global_config = config.model.global_config
# TriangleAttentionTest(args, config, global_config, is_training=True)
# TriangleMultiplicationTest(args, config, global_config, is_training=True)
# OuterProductMeanTest(args, config, global_config, is_training=True)
# TransitionTest(args, config, global_config, is_training=True)
InvariantPointAttentionTest(args, config, global_config)
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Train deep protein docking')
parser.add_argument('-fasta_path', default='T1024.fas', type=str)
parser.add_argument('-output_dir',
default='/media/lupoglaz/AlphaFold2Output',
type=str)
parser.add_argument('-model_name', default='model_1', type=str)
parser.add_argument('-data_dir',
default='/media/lupoglaz/AlphaFold2Data',
type=str)
args = parser.parse_args()
model_config = model_config(args.model_name)
model_config.data.eval.num_ensemble = 1
model_config.data.common.use_templates = False
af2features = AlphaFoldFeatures(config=model_config)
features_path = Path(
args.output_dir) / Path('T1024') / Path('features.pkl')
proc_features_path = Path(
args.output_dir) / Path('T1024') / Path('proc_features.pkl')
with open(features_path, 'rb') as f:
raw_feature_dict = pickle.load(f)
with open(proc_features_path, 'rb') as f:
af2_proc_features = pickle.load(f)
this_proc_features = af2features(raw_feature_dict, random_seed=42)
