Merge pull request #2151 from deepchem/gcsave

class MultitaskFitTransformRegressor(MultitaskRegressor):

  """Implements a MultitaskRegressor that performs on-the-fly transformation during fit/predict.

  Example:

  Examples

  --------

  >>> n_samples = 10

  >>> n_features = 3

  >>> n_tasks = 1

      returns the values of the uncertainty outputs.

    other_output_types: list, optional

      Provides a list of other output_types (strings) to predict from model.

    Returns:

    Returns

    -------

    a NumPy array of the model produces a single output, or a list of arrays

    if it produces multiple outputs

    """

    if it produces multiple outputs

    """

    generator = self.default_generator(

        dataset, mode='predict', pad_batches=False)

        dataset, mode='predict', deterministic=True, pad_batches=False)

    return self.predict_on_generator(

        generator,

        transformers=transformers,

def test_skewed_missing_classification_overfit():

  """TG, skewed data, few actives

  """MultitaskClassifier, skewed data, few actives

  Test MultitaskClassifier overfit 0/1 datasets with missing data and few

  actives. This is intended to be as close to singletask MUV datasets as

  assert scores[classification_metric.name] > .9

def test_tf_robust_multitask_classification_overfit():

  """Test tf robust multitask overfits tiny data."""

def test_robust_multitask_classification_overfit():

  """Test robust multitask overfits tiny data."""

  n_tasks = 10

  n_samples = 10

  n_features = 3

We still try to run them regularly, so hopefully the problem will be discovered

fairly soon.

Testing Machine Learning Models

-------------------------------

Testing the correctness of a machine learning model can be quite

tricky to do in practice. When adding a new machine learning model to

DeepChem, you should add at least a few basic types of unit tests:

- Overfitting test: Create a small synthetic dataset and test that

your model can learn this datasest with high accuracy. For regression

and classification task, this should correspond to low training error

on the dataset. For generative tasks, this should correspond to low

training loss on the dataset.

- Reloading test: Check that a trained model can be saved to disk and

reloaded correctly. This should involve checking that predictions from

the saved and reloaded models matching exactly.

Note that unit tests are not sufficient to gauge the real performance

of a model. You should benchmark your model on larger datasets as well

and report your benchmarking tests in the PR comments.

Type Annotations

----------------