classifier test overfit

    scores = model.evaluate(dataset, [classification_metric])

    assert scores[classification_metric.name] > .9

  def test_IRV_multitask_classification_overfit(self):

    """Test IRV classifier overfits tiny data."""

    n_tasks = 5

    n_samples = 10

    n_features = 128

    n_classes = 2

    # Generate dummy dataset

    np.random.seed(123)

    ids = np.arange(n_samples)

    X = np.random.randint(2, size=(n_samples, n_features))

    y = np.zeros((n_samples, n_tasks))

    w = np.ones((n_samples, n_tasks))

    dataset = dc.data.NumpyDataset(X, y, w, ids)

    IRV_transformer = dc.trans.IRVTransformer(5, n_tasks, dataset)

    dataset_trans = IRV_transformer.transform(dataset)

    classification_metric = dc.metrics.Metric(

        dc.metrics.accuracy_score, task_averager=np.mean)

    model = dc.models.TensorflowMultiTaskIRVClassifier(

        n_tasks, K=5, learning_rate=0.001, batch_size=n_samples)

    # Fit trained model

    model.fit(dataset_trans)

    model.save()

    # Eval model on train

    scores = model.evaluate(dataset_trans, [classification_metric])

    assert scores[classification_metric.name] > .9

  def test_sklearn_multitask_regression_overfit(self):

    """Test SKLearn singletask-to-multitask overfits tiny regression data."""

    n_tasks = 2

    sims = sorted(sims, reverse=True)

    IRV_transformer = dc.trans.IRVTransformer(10, n_tasks, dataset)

    test_dataset_trans = IRV_transformer.transform(test_dataset)

    dataset_trans = IRV_transformer.transform(dataset)

    assert test_dataset_trans.X.shape == (test_samples, 20*n_tasks)

    assert np.allclose(test_dataset_trans.X[0,:10], sims[:10])

    assert np.allclose(test_dataset_trans.X[0,10:20], [0]*10)

    assert not np.isclose(dataset_trans.X[0,0], 1.)