some minor cleanup / corrections

Some examples of training models on the SAMPL(FreeSolv) dataset included in :code:`dc.molnet`.

First we'll load our tasks, dataset  splits, and transformers from molnet:

We'll be using its ``smiles`` field to train models to predict its experimentally measured solvation energy (``expt``).

First, we'll load our libraries:

.. doctest::

   >>> import numpy as np

   >>> # Load SAMPL dataset

   >>> SAMPL_tasks, SAMPL_datasets, transformers = load_sampl()

   >>> SAMPL_tasks

   ['expt']

   >>> train_dataset, valid_dataset, test_dataset = SAMPL_datasets

   >>>

   >>> # We'll train a multitask regressor (fully connected network)

   ...     learning_rate=0.001,

   ...     batch_size=50)

   >>> 

   >>> # Fit trained model

   >>> # Fit trained model (returns average loss over the most recent checkpoint interval)

   >>> model.fit(train_dataset)

   0.1726440668106079

   >>> 

   >>> np.random.seed(123)

   >>> tf.random.set_seed(123)

   >>> # Load SAMPL dataset

   >>> SAMPL_tasks, SAMPL_datasets, transformers = dc.molnet.load_sampl(

   >>> SAMPL_tasks, SAMPL_datasets, transformers = load_sampl(

   ...     featurizer='GraphConv')

   >>> train_dataset, valid_dataset, test_dataset = SAMPL_datasets

   >>>

   >>> model = dc.models.GraphConvModel(len(SAMPL_tasks), mode='regression')

   >>> 

   >>> # Fit trained model

   >>> # Fit trained model (returns average loss over the most recent checkpoint interval)

   >>> model.fit(train_dataset, nb_epoch=20)

   0.05753047466278076

   >>>