Аtom_target prediction

[SergeiNikolenko]

I see in your part responsible for the implementation of prediction of atomic features, but if you just specify target column with them, the model does not work and gives the error
ValueError: Unrecognized targets of column CDD in ../data/QM_100.csv. there is a one-dimensional array there.

[kevingreenman]

Hi @SergeiNikolenko, thanks for your question. This isn't enough information for us to understand and reproduce your error. If you can provide the command you ran and the full error traceback (and preferably also a minimal working example with data), then we can help debug this.

[SergeiNikolenko]

I start the workout with this command:

arguments = [
    '--data_path', 'demo.csv',
    '--dataset_type', 'regression',
    '--save_dir', 'test_checkpoints_reg',
    '--smiles_columns', 'smiles',
    '--target_columns', 'CDD',
    '--epochs', '100',
    '--save_smiles_splits',
    '--adding_h',
    '--show_individual_scores',

]

args = chemprop.args.TrainArgs().parse_args(arguments)
mean_score, std_score = chemprop.train.cross_validate(args=args, train_func=chemprop.train.run_training)

This error occurs:

-------------------
ValueError                                Traceback (most recent call last)
/home/nikolenko/work/chemprop/regression.ipynb Cell 5 line 1
      <a href='vscode-notebook-cell://ssh-remote%2Bserv4/home/nikolenko/work/chemprop/regression.ipynb#W5sdnNjb2RlLXJlbW90ZQ%3D%3D?line=0'>1</a> arguments = [
      <a href='vscode-notebook-cell://ssh-remote%2Bserv4/home/nikolenko/work/chemprop/regression.ipynb#W5sdnNjb2RlLXJlbW90ZQ%3D%3D?line=1'>2</a>     '--data_path', 'demo.csv',
      <a href='vscode-notebook-cell://ssh-remote%2Bserv4/home/nikolenko/work/chemprop/regression.ipynb#W5sdnNjb2RlLXJlbW90ZQ%3D%3D?line=2'>3</a>     '--dataset_type', 'regression',
   (...)
     <a href='vscode-notebook-cell://ssh-remote%2Bserv4/home/nikolenko/work/chemprop/regression.ipynb#W5sdnNjb2RlLXJlbW90ZQ%3D%3D?line=10'>11</a> 
     <a href='vscode-notebook-cell://ssh-remote%2Bserv4/home/nikolenko/work/chemprop/regression.ipynb#W5sdnNjb2RlLXJlbW90ZQ%3D%3D?line=11'>12</a> ]
     <a href='vscode-notebook-cell://ssh-remote%2Bserv4/home/nikolenko/work/chemprop/regression.ipynb#W5sdnNjb2RlLXJlbW90ZQ%3D%3D?line=13'>14</a> args = chemprop.args.TrainArgs().parse_args(arguments)
---> <a href='vscode-notebook-cell://ssh-remote%2Bserv4/home/nikolenko/work/chemprop/regression.ipynb#W5sdnNjb2RlLXJlbW90ZQ%3D%3D?line=14'>15</a> mean_score, std_score = chemprop.train.cross_validate(args=args, train_func=chemprop.train.run_training)

File ~/work/chemprop/chemprop/utils.py:590, in timeit.<locals>.timeit_decorator.<locals>.wrap(*args, **kwargs)
    587 @wraps(func)
    588 def wrap(*args, **kwargs) -> Any:
    589     start_time = time()
--> 590     result = func(*args, **kwargs)
    591     delta = timedelta(seconds=round(time() - start_time))
    592     info = logging.getLogger(logger_name).info if logger_name is not None else print

File ~/work/chemprop/chemprop/train/cross_validate.py:76, in cross_validate(args, train_func)
     74 # Get data
     75 debug('Loading data')
---> 76 data = get_data(
     77     path=args.data_path,
     78     args=args,
     79     logger=logger,
     80     skip_none_targets=True,
     81     data_weights_path=args.data_weights_path
     82 )
     83 validate_dataset_type(data, dataset_type=args.dataset_type)
     84 args.features_size = data.features_size()

File ~/work/chemprop/chemprop/data/utils.py:513, in get_data(path, smiles_columns, target_columns, ignore_columns, skip_invalid_smiles, args, data_weights_path, features_path, features_generator, phase_features_path, atom_descriptors_path, bond_descriptors_path, constraints_path, max_data_size, store_row, logger, loss_function, skip_none_targets)
    511         targets.append(np.array(bond_target_arranged))
    512     else:
--> 513         raise ValueError(f'Unrecognized targets of column {column} in {path}.')
    514 else:
    515     targets.append(float(value))

ValueError: Unrecognized targets of column CDD in demo.csv. 

here's the file I'm running with demo.csv

[kevingreenman]

It looks like you're missing the argument --is_atom_bond_targets

[SergeiNikolenko]

Thank you very much, I've been trying to implement atomic property prediction for a long time and I haven't seen explicit mention in the documentation how to do it.

[kevingreenman]

This feature is documented here. Our Read The Docs page is a couple years out of date if that's where you were looking previously. The README is the best place to look for documentation right now. The descriptions of arguments in args.py may also be helpful. The Read The Docs will be brought up to date with the imminent release of Chemprop v2.0. However, for your purposes, note that atom and bond targets is not planned until v2.2, so you may need to continue using v1.6 (current release) or v1.7 (about to be released).

[SergeiNikolenko]

Yeah, I've recently noticed that the documentation is outdated, and it's better to usу README