ValueError: mode must be either 'classification' or 'regression' in Hyper-parameter optimization

Hi, thank you for sharing your great work.

I am trying hyper-parameter optimization:

import deepchem as dc

tasks, datasets, transformers = dc.molnet.load_delaney(featurizer='GraphConv')
train_dataset, valid_dataset, test_dataset = datasets

metric = dc.metrics.Metric(dc.metrics.r2_score)

params_dict = {
    'n_tasks': [len(tasks)],
    'mode': 'regression',
    'graph_conv_layers': [[64], [64, 64]],
    'dropouts': [0.2, 0.5]
}

optimizer = dc.hyper.GridHyperparamOpt(dc.models.GraphConvModel)

best_model, best_hyperparams, all_results = optimizer.hyperparam_search(
        params_dict, train_dataset, valid_dataset, metric, transformers)

However, I get the error:

ValueError: mode must be either 'classification' or 'regression'

Environment:

• OS: macOS 11.5
• Python 3.7
• deepchem == 2.5.0
• TensorFlow == 2.6.0
• PyTorch == 1.7.0

Thank you for your help.

[rbharath]

Looking at the source, I’m able to confirm it does indeed return a tuple and not a dictionary:

https://github.com/deepchem/deepchem/blob/master/deepchem/hyper/grid_search.py#L172

We should correct this. It’s a little delicate since we don’t want to break backwards compatibility in anyone’s code that’s been assuming tuples instead of dictionaries. I suggest for now we just update the documentation so it’s clear for users and decide later whether to make a breaking fix

[arunppsg]

Docs fixes in #2712