Change radius to 2

  def predict_mols(self, mols):

    featurizer = CircularFingerprint(

        size=self.n_features, radius=3, chiral=True)

        size=self.n_features, radius=2, chiral=True)

    features = np.expand_dims(featurizer.featurize(mols), axis=1)

    features = np.concatenate([features, features], axis=1)

    ds = NumpyDataset(features, None, None, None)

    for layer, column in zip([self.m1_features, self.m2_features],

                             feature_columns):

      tensors[layer] = tf.feature_column.input_layer(features, [column])

    if weight_column is not None:

      tensors[self.task_weights[0]] = tf.feature_column.input_layer(

          features, [weight_column])

    if labels is not None:

      tensors[self.labels[0]] = tf.cast(labels, tf.int32)

    return tensors

    x_col1 = tf.feature_column.numeric_column('x1', shape=(n_features,))

    x_col2 = tf.feature_column.numeric_column('x2', shape=(n_features,))

    weight_col = tf.feature_column.numeric_column('weights', shape=(1,))

    def accuracy(labels, predictions, weights):

      return tf.metrics.accuracy(labels, tf.round(predictions), weights)

    metrics = {'accuracy': accuracy}

    estimator = model.make_estimator(

        feature_columns=[x_col1, x_col2], metrics=metrics)

        feature_columns=[x_col1, x_col2], metrics={}, weight_column=weight_col)

    # Train the model.

    # Evaluate the model.

    results = estimator.evaluate(input_fn=lambda: input_fn(1))

    print(results)

    assert results['loss'] < 1e-4

    # TODO(LESWING) Discuss with peastman.

    #  The output here is human readable

    # score 1-5 per molecule not a probability of class

    # assert results['accuracy'] > 0.9