Another attempt at getting test to pass on travis

    """Test fitting a KerasModel defined as a graph."""

    n_data_points = 10

    n_features = 2

    np.random.seed(1234)

    X = np.random.rand(n_data_points, n_features)

    y = (X[:, 0] > X[:, 1]).astype(np.float32)

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

        keras_model,

        dc.models.losses.SigmoidCrossEntropy(),

        output_types=['prediction', 'loss'],

        learning_rate=0.004)

        learning_rate=0.005)

    model.fit(dataset, nb_epoch=1000)

    prediction = np.squeeze(model.predict_on_batch(X))

    print(y)