Python tensorflow.py_func() Examples

def _py_func_with_gradient(func, inp, Tout, stateful=True, name=None,
                           grad_func=None):
    """
    PyFunc defined as given by Tensorflow
    :param func: Custom Function
    :param inp: Function Inputs
    :param Tout: Ouput Type of out Custom Function
    :param stateful: Calculate Gradients when stateful is True
    :param name: Name of the PyFunction
    :param grad: Custom Gradient Function
    :return:
    """
    # Generate random name in order to avoid conflicts with inbuilt names
    rnd_name = 'PyFuncGrad-' + '%0x' % getrandbits(30 * 4)

    # Register Tensorflow Gradient
    tf.RegisterGradient(rnd_name)(grad_func)

    # Get current graph
    g = tf.get_default_graph()

    # Add gradient override map
    with g.gradient_override_map(
            {"PyFunc": rnd_name, "PyFuncStateless": rnd_name}):
        return tf.py_func(func, inp, Tout, stateful=stateful, name=name) 

def compute_gradients(self, loss, var_list, **kwargs):
        grads_and_vars = tf.train.AdamOptimizer.compute_gradients(self, loss, var_list, **kwargs)
        grads_and_vars = [(g, v) for g, v in grads_and_vars if g is not None]
        flat_grad = tf.concat([tf.reshape(g, (-1,)) for g, v in grads_and_vars], axis=0)
        shapes = [v.shape.as_list() for g, v in grads_and_vars]
        sizes = [int(np.prod(s)) for s in shapes]

        num_tasks = self.comm.Get_size()
        buf = np.zeros(sum(sizes), np.float32)

        def _collect_grads(flat_grad):
            self.comm.Allreduce(flat_grad, buf, op=MPI.SUM)
            np.divide(buf, float(num_tasks), out=buf)
            return buf

        avg_flat_grad = tf.py_func(_collect_grads, [flat_grad], tf.float32)
        avg_flat_grad.set_shape(flat_grad.shape)
        avg_grads = tf.split(avg_flat_grad, sizes, axis=0)
        avg_grads_and_vars = [(tf.reshape(g, v.shape), v)
                    for g, (_, v) in zip(avg_grads, grads_and_vars)]

        return avg_grads_and_vars 

def iou_rotate_calculate(boxes1, boxes2, use_gpu=True, gpu_id=0):
    '''

    :param boxes_list1:[N, 8] tensor
    :param boxes_list2: [M, 8] tensor
    :return:
    '''

    boxes1 = tf.cast(boxes1, tf.float32)
    boxes2 = tf.cast(boxes2, tf.float32)
    if use_gpu:

        iou_matrix = tf.py_func(rbbx_overlaps,
                                inp=[boxes1, boxes2, gpu_id],
                                Tout=tf.float32)
    else:
        iou_matrix = tf.py_func(get_iou_matrix, inp=[boxes1, boxes2],
                                Tout=tf.float32)

    iou_matrix = tf.reshape(iou_matrix, [tf.shape(boxes1)[0], tf.shape(boxes2)[0]])

    return iou_matrix 

def nms_rotate_gpu(boxes_list, scores, iou_threshold, use_angle_condition=False, angle_gap_threshold=0, device_id=0):
    if use_angle_condition:
        x_c, y_c, w, h, theta = tf.unstack(boxes_list, axis=1)
        boxes_list = tf.transpose(tf.stack([x_c, y_c, w, h, theta]))
        det_tensor = tf.concat([boxes_list, tf.expand_dims(scores, axis=1)], axis=1)
        keep = tf.py_func(rotate_gpu_nms,
                          inp=[det_tensor, iou_threshold, device_id],
                          Tout=tf.int64)
        return keep
    else:
        x_c, y_c, w, h, theta = tf.unstack(boxes_list, axis=1)
        boxes_list = tf.transpose(tf.stack([x_c, y_c, w, h, theta]))
        det_tensor = tf.concat([boxes_list, tf.expand_dims(scores, axis=1)], axis=1)
        keep = tf.py_func(rotate_gpu_nms,
                          inp=[det_tensor, iou_threshold, device_id],
                          Tout=tf.int64)
        keep = tf.reshape(keep, [-1])
        return keep 

def simulate(self, action):
    """Step the batch of environments.

    The results of the step can be accessed from the variables defined below.

    Args:
      action: Tensor holding the batch of actions to apply.

    Returns:
      Operation.
    """
    with tf.name_scope('environment/simulate'):
      if action.dtype in (tf.float16, tf.float32, tf.float64):
        action = tf.check_numerics(action, 'action')
      observ_dtype = self._parse_dtype(self._batch_env.observation_space)
      observ, reward, done = tf.py_func(
          lambda a: self._batch_env.step(a)[:3], [action],
          [observ_dtype, tf.float32, tf.bool], name='step')
      observ = tf.check_numerics(observ, 'observ')
      reward = tf.check_numerics(reward, 'reward')
      return tf.group(
          self._observ.assign(observ),
          self._action.assign(action),
          self._reward.assign(reward),
          self._done.assign(done)) 

def simulate(self, action):
    """Step the batch of environments.

    The results of the step can be accessed from the variables defined below.

    Args:
      action: Tensor holding the batch of actions to apply.

    Returns:
      Operation.
    """
    with tf.name_scope('environment/simulate'):
      if action.dtype in (tf.float16, tf.float32, tf.float64):
        action = tf.check_numerics(action, 'action')
      observ_dtype = self._parse_dtype(self._batch_env.observation_space)
      observ, reward, done = tf.py_func(
          lambda a: self._batch_env.step(a)[:3], [action],
          [observ_dtype, tf.float32, tf.bool], name='step')
      observ = tf.check_numerics(observ, 'observ')
      reward = tf.check_numerics(reward, 'reward')
      return tf.group(
          self._observ.assign(observ),
          self._action.assign(action),
          self._reward.assign(reward),
          self._done.assign(done)) 

def reset(self, indices=None):
    """Reset the batch of environments.

    Args:
      indices: The batch indices of the environments to reset; defaults to all.

    Returns:
      Batch tensor of the new observations.
    """
    if indices is None:
      indices = tf.range(len(self._batch_env))
    observ_dtype = self._parse_dtype(self._batch_env.observation_space)
    observ = tf.py_func(
        self._batch_env.reset, [indices], observ_dtype, name='reset')
    observ = tf.check_numerics(observ, 'observ')
    reward = tf.zeros_like(indices, tf.float32)
    done = tf.zeros_like(indices, tf.bool)
    with tf.control_dependencies([
        tf.scatter_update(self._observ, indices, observ),
        tf.scatter_update(self._reward, indices, reward),
        tf.scatter_update(self._done, indices, done)]):
      return tf.identity(observ) 

def proposal_target_layer(self, input, classes, name):
        if isinstance(input[0], tuple):
            input[0] = input[0][0]
        with tf.variable_scope(name) as scope:
            #inputs: 'rpn_rois','gt_boxes', 'gt_ishard', 'dontcare_areas'
            rois,labels,bbox_targets,bbox_inside_weights,bbox_outside_weights \
                = tf.py_func(proposal_target_layer_py,
                             [input[0],input[1],input[2],input[3],classes],
                             [tf.float32,tf.float32,tf.float32,tf.float32,tf.float32])
            # rois <- (1 x H x W x A, 5) e.g. [0, x1, y1, x2, y2]
            # rois = tf.convert_to_tensor(rois, name='rois')
            rois = tf.reshape(rois, [-1, 5], name='rois') # goes to roi_pooling
            labels = tf.convert_to_tensor(tf.cast(labels,tf.int32), name = 'labels') # goes to FRCNN loss
            bbox_targets = tf.convert_to_tensor(bbox_targets, name = 'bbox_targets') # goes to FRCNN loss
            bbox_inside_weights = tf.convert_to_tensor(bbox_inside_weights, name = 'bbox_inside_weights')
            bbox_outside_weights = tf.convert_to_tensor(bbox_outside_weights, name = 'bbox_outside_weights')

            self.layers['rois'] = rois

            return rois, labels, bbox_targets, bbox_inside_weights, bbox_outside_weights 

def anchor_target_layer(self, input, _feat_stride, anchor_scales, name):
        if isinstance(input[0], tuple):
            input[0] = input[0][0]

        with tf.variable_scope(name) as scope:
            # 'rpn_cls_score', 'gt_boxes', 'gt_ishard', 'dontcare_areas', 'im_info'
            rpn_labels,rpn_bbox_targets,rpn_bbox_inside_weights,rpn_bbox_outside_weights = \
                tf.py_func(anchor_target_layer_py,
                           [input[0],input[1],input[2],input[3],input[4], _feat_stride, anchor_scales],
                           [tf.float32,tf.float32,tf.float32,tf.float32])

            rpn_labels = tf.convert_to_tensor(tf.cast(rpn_labels,tf.int32), name = 'rpn_labels') # shape is (1 x H x W x A, 2)
            rpn_bbox_targets = tf.convert_to_tensor(rpn_bbox_targets, name = 'rpn_bbox_targets') # shape is (1 x H x W x A, 4)
            rpn_bbox_inside_weights = tf.convert_to_tensor(rpn_bbox_inside_weights , name = 'rpn_bbox_inside_weights') # shape is (1 x H x W x A, 4)
            rpn_bbox_outside_weights = tf.convert_to_tensor(rpn_bbox_outside_weights , name = 'rpn_bbox_outside_weights') # shape is (1 x H x W x A, 4)


            return rpn_labels, rpn_bbox_targets, rpn_bbox_inside_weights, rpn_bbox_outside_weights 

def simulate(self, action):
    """Step the batch of environments.

    The results of the step can be accessed from the variables defined below.

    Args:
      action: Tensor holding the batch of actions to apply.

    Returns:
      Operation.
    """
    with tf.name_scope('environment/simulate'):
      if action.dtype in (tf.float16, tf.float32, tf.float64):
        action = tf.check_numerics(action, 'action')
      observ_dtype = utils.parse_dtype(self._batch_env.observation_space)
      observ, reward, done = tf.py_func(
          lambda a: self._batch_env.step(a)[:3], [action],
          [observ_dtype, tf.float32, tf.bool], name='step')
      observ = tf.check_numerics(observ, 'observ')
      reward = tf.check_numerics(reward, 'reward')
      reward.set_shape((len(self),))
      done.set_shape((len(self),))
      with tf.control_dependencies([self._observ.assign(observ)]):
        return tf.identity(reward), tf.identity(done) 

def rouge_l_fscore(predictions, labels, **unused_kwargs):
  """ROUGE scores computation between labels and predictions.

  This is an approximate ROUGE scoring method since we do not glue word pieces
  or decode the ids and tokenize the output.

  Args:
    predictions: tensor, model predictions
    labels: tensor, gold output.

  Returns:
    rouge_l_fscore: approx rouge-l f1 score.
  """
  outputs = tf.to_int32(tf.argmax(predictions, axis=-1))
  # Convert the outputs and labels to a [batch_size, input_length] tensor.
  outputs = tf.squeeze(outputs, axis=[-1, -2])
  labels = tf.squeeze(labels, axis=[-1, -2])
  rouge_l_f_score = tf.py_func(rouge_l_sentence_level, (outputs, labels),
                               tf.float32)
  return rouge_l_f_score, tf.constant(1.0) 

def rouge_2_fscore(predictions, labels, **unused_kwargs):
  """ROUGE-2 F1 score computation between labels and predictions.

  This is an approximate ROUGE scoring method since we do not glue word pieces
  or decode the ids and tokenize the output.

  Args:
    predictions: tensor, model predictions
    labels: tensor, gold output.

  Returns:
    rouge2_fscore: approx rouge-2 f1 score.
  """

  outputs = tf.to_int32(tf.argmax(predictions, axis=-1))
  # Convert the outputs and labels to a [batch_size, input_length] tensor.
  outputs = tf.squeeze(outputs, axis=[-1, -2])
  labels = tf.squeeze(labels, axis=[-1, -2])
  rouge_2_f_score = tf.py_func(rouge_n, (outputs, labels), tf.float32)
  return rouge_2_f_score, tf.constant(1.0) 

def add_noise(ids, sequence_length):
  """Wraps add_noise_python for a batch of tensors."""

  def _add_noise_single(ids, sequence_length):
    noisy_ids = add_noise_python(ids[:sequence_length])
    noisy_sequence_length = len(noisy_ids)
    ids[:noisy_sequence_length] = noisy_ids
    ids[noisy_sequence_length:] = 0
    return ids, np.int32(noisy_sequence_length)

  noisy_ids, noisy_sequence_length = tf.map_fn(
      lambda x: tf.py_func(_add_noise_single, x, [ids.dtype, tf.int32]),
      [ids, sequence_length],
      dtype=[ids.dtype, tf.int32],
      back_prop=False)

  noisy_ids.set_shape(ids.get_shape())
  noisy_sequence_length.set_shape(sequence_length.get_shape())

  return noisy_ids, noisy_sequence_length


# Step 3 

def _reset_non_empty(self, indices):
    """Reset the batch of environments.

    Args:
      indices: The batch indices of the environments to reset; defaults to all.

    Returns:
      Batch tensor of the new observations.
    """
    observ_dtype = utils.parse_dtype(self._batch_env.observation_space)
    observ = tf.py_func(
        self._batch_env.reset, [indices], observ_dtype, name='reset')
    observ = tf.check_numerics(observ, 'observ')
    with tf.control_dependencies([
        tf.scatter_update(self._observ, indices, observ)]):
      return tf.identity(observ) 

def compute_gradients(self, loss, var_list, **kwargs):
        grads_and_vars = tf.train.AdamOptimizer.compute_gradients(self, loss, var_list, **kwargs)
        grads_and_vars = [(g, v) for g, v in grads_and_vars if g is not None]
        flat_grad = tf.concat([tf.reshape(g, (-1,)) for g, v in grads_and_vars], axis=0)
        shapes = [v.shape.as_list() for g, v in grads_and_vars]
        sizes = [int(np.prod(s)) for s in shapes]

        num_tasks = self.comm.Get_size()
        buf = np.zeros(sum(sizes), np.float32)

        def _collect_grads(flat_grad):
            self.comm.Allreduce(flat_grad, buf, op=MPI.SUM)
            np.divide(buf, float(num_tasks), out=buf)
            return buf

        avg_flat_grad = tf.py_func(_collect_grads, [flat_grad], tf.float32)
        avg_flat_grad.set_shape(flat_grad.shape)
        avg_grads = tf.split(avg_flat_grad, sizes, axis=0)
        avg_grads_and_vars = [(tf.reshape(g, v.shape), v)
                    for g, (_, v) in zip(avg_grads, grads_and_vars)]

        return avg_grads_and_vars 

def compute_gradients(self, loss, var_list, **kwargs):
        grads_and_vars = tf.train.AdamOptimizer.compute_gradients(self, loss, var_list, **kwargs)
        grads_and_vars = [(g, v) for g, v in grads_and_vars if g is not None]
        flat_grad = tf.concat([tf.reshape(g, (-1,)) for g, v in grads_and_vars], axis=0)
        shapes = [v.shape.as_list() for g, v in grads_and_vars]
        sizes = [int(np.prod(s)) for s in shapes]

        num_tasks = self.comm.Get_size()
        buf = np.zeros(sum(sizes), np.float32)

        def _collect_grads(flat_grad):
            self.comm.Allreduce(flat_grad, buf, op=MPI.SUM)
            np.divide(buf, float(num_tasks), out=buf)
            return buf

        avg_flat_grad = tf.py_func(_collect_grads, [flat_grad], tf.float32)
        avg_flat_grad.set_shape(flat_grad.shape)
        avg_grads = tf.split(avg_flat_grad, sizes, axis=0)
        avg_grads_and_vars = [(tf.reshape(g, v.shape), v)
                    for g, (_, v) in zip(avg_grads, grads_and_vars)]

        return avg_grads_and_vars 

def generate_gt_graph(gt_quadrilaterals, input_gt_class_ids, image_shape, width_stride, max_gt_num):
    """

    :param gt_quadrilaterals: [mat_gt_num,(x1,y1,x2,y2,x3,y3,x4,y4,tag)] 左上、右上、右下、左下(顺时针)
    :param input_gt_class_ids:
    :param image_shape:
    :param width_stride:
    :param max_gt_num:
    :return:
    """

    gt_quadrilaterals = tf_utils.remove_pad(gt_quadrilaterals)
    input_gt_class_ids = tf_utils.remove_pad(input_gt_class_ids)
    gt_boxes, gt_class_ids = tf.py_func(func=gt_utils.gen_gt_from_quadrilaterals,
                          inp=[gt_quadrilaterals, input_gt_class_ids, image_shape, width_stride],
                          Tout=[tf.float32] * 2)
    return tf_utils.pad_list_to_fixed_size([gt_boxes, gt_class_ids], max_gt_num) 

def rouge_2_fscore(predictions, labels, **unused_kwargs):
  """ROUGE-2 F1 score computation between labels and predictions.

  This is an approximate ROUGE scoring method since we do not glue word pieces
  or decode the ids and tokenize the output.

  Args:
    predictions: tensor, model predictions (batch_size, <=max_dec_steps)
    labels: tensor, gold output. (batch_size, max_dec_steps)

  Returns:
    rouge2_fscore: approx rouge-2 f1 score.
  """

  rouge_2_f_score = tf.py_func(rouge_n, (predictions, labels), [tf.float32])
  return rouge_2_f_score, tf.constant(1.0) 

def rouge_l_fscore(hypothesis, references, **unused_kwargs):
  """ROUGE scores computation between labels and predictions.

  This is an approximate ROUGE scoring method since we do not glue word pieces
  or decode the ids and tokenize the output.

  Args:
    predictions: tensor, model predictions (batch_size, <=max_dec_steps)
    labels: tensor, gold output. (batch_size, max_dec_steps)

  Returns:
    rouge_l_fscore: approx rouge-l f1 score.
  """
  rouge_l_f_score = tf.py_func(rouge_l_sentence_level, (hypothesis, references), [tf.float32])
  #rouge_l_f_score = tf.py_func(rouge_l_sentence_level, (hypothesis, references), tf.float32)
  return rouge_l_f_score 

def anchor_target_layer(cls_pred, bbox, im_info, scope_name):
    with tf.variable_scope(scope_name) as scope:
        # 'rpn_cls_score', 'gt_boxes', 'im_info'
        rpn_labels, rpn_bbox_targets, rpn_bbox_inside_weights, rpn_bbox_outside_weights = \
            tf.py_func(anchor_target_layer_py,
                       [cls_pred, bbox, im_info, [16, ], [16]],
                       [tf.float32, tf.float32, tf.float32, tf.float32])

        rpn_labels = tf.convert_to_tensor(tf.cast(rpn_labels, tf.int32),
                                          name='rpn_labels')
        rpn_bbox_targets = tf.convert_to_tensor(rpn_bbox_targets,
                                                name='rpn_bbox_targets')
        rpn_bbox_inside_weights = tf.convert_to_tensor(rpn_bbox_inside_weights,
                                                       name='rpn_bbox_inside_weights')
        rpn_bbox_outside_weights = tf.convert_to_tensor(rpn_bbox_outside_weights,
                                                        name='rpn_bbox_outside_weights')

        return [rpn_labels, rpn_bbox_targets, rpn_bbox_inside_weights, rpn_bbox_outside_weights] 

def tf_msssim_np(img1, img2, data_format='NHWC'):
    assert img1.shape.ndims == img2.shape.ndims == 4, 'Expected {}, got {}'.format(data_format, img1.shape, img2.shape)
    assert tf.uint8.is_compatible_with(img1.dtype), 'Expected uint8 intput'
    assert tf.uint8.is_compatible_with(img2.dtype), 'Expected uint8 intput'

    if data_format == 'NCHW':
        def make_NHWC(x):
            return tf.transpose(x, (0, 2, 3, 1), name='make_NHWC')
        return tf_msssim_np(make_NHWC(img1), make_NHWC(img2), data_format='NHWC')

    assert img1.shape[3] == 3, 'Expected 3-channel images, got {}'.format(img1)

    with tf.name_scope('ms-ssim_np'):
        v = tf.py_func(_calc_msssim_orig, [img1, img2], tf.float32, stateful=False, name='MS-SSIM')
        v.set_shape(())
        return v 

def compute_gradients(self, loss, var_list, **kwargs):
        grads_and_vars = tf.train.AdamOptimizer.compute_gradients(self, loss, var_list, **kwargs)
        grads_and_vars = [(g, v) for g, v in grads_and_vars if g is not None]
        flat_grad = tf.concat([tf.reshape(g, (-1,)) for g, v in grads_and_vars], axis=0)
        shapes = [v.shape.as_list() for g, v in grads_and_vars]
        sizes = [int(np.prod(s)) for s in shapes]

        num_tasks = self.comm.Get_size()
        buf = np.zeros(sum(sizes), np.float32)

        def _collect_grads(flat_grad):
            self.comm.Allreduce(flat_grad, buf, op=MPI.SUM)
            np.divide(buf, float(num_tasks), out=buf)
            return buf

        avg_flat_grad = tf.py_func(_collect_grads, [flat_grad], tf.float32)
        avg_flat_grad.set_shape(flat_grad.shape)
        avg_grads = tf.split(avg_flat_grad, sizes, axis=0)
        avg_grads_and_vars = [(tf.reshape(g, v.shape), v)
                    for g, (_, v) in zip(avg_grads, grads_and_vars)]

        return avg_grads_and_vars 

def bleu_score(predictions, labels, **unused_kwargs):
  """BLEU score computation between labels and predictions.

  An approximate BLEU scoring method since we do not glue word pieces or
  decode the ids and tokenize the output. By default, we use ngram order of 4
  and use brevity penalty. Also, this does not have beam search.

  Args:
    predictions: tensor, model predictions
    labels: tensor, gold output.

  Returns:
    bleu: int, approx bleu score
  """
  outputs = tf.to_int32(tf.argmax(predictions, axis=-1))
  # Convert the outputs and labels to a [batch_size, input_length] tensor.
  outputs = tf.squeeze(outputs, axis=[-1, -2])
  labels = tf.squeeze(labels, axis=[-1, -2])

  bleu = tf.py_func(compute_bleu, (labels, outputs), tf.float32)
  return bleu, tf.constant(1.0) 

def mpi_bcast(comm, value, root=0):
    """Broadcast value from root to other processes via a TensorFlow py_func."""
    value = tf.convert_to_tensor(value)
    if comm.Get_size() == 1:
        return value
    comm = comm.Dup()  # Allow parallelism at graph execution time
    if comm.Get_rank() == root:
        out = tf.py_func(partial(comm.bcast, root=root), [value], value.dtype)
    else:
        out = tf.py_func(partial(comm.bcast, None, root=root), [], value.dtype)
    out.set_shape(value.shape)
    return out 

def loss(gt, outputs, class_weights, bboxes_weights):
  gt_classes, gt_bboxes, gt_mask = gt
  feat_classes = outputs[0]
  feat_bboxes = outputs[1]

  gt_mask = tf.reshape(gt_mask, [-1])
  logits_classes = tf.reshape(feat_classes, [-1, tf.shape(feat_classes)[2]])
  gt_classes = tf.reshape(gt_classes, [-1, 1])
  logits_bboxes = tf.reshape(feat_bboxes, [-1, 4])
  gt_bboxes = tf.reshape(gt_bboxes, [-1, 4])

  # # heuristic sampling
  # fg_index, bg_index = tf.py_func(
  #   heuristic_sampling, [gt_mask], (tf.int64, tf.int64))

  # hard negative mining
  fg_index, bg_index = hard_negative_mining(logits_classes, gt_mask)

  loss_classes = class_weights * create_loss_classes_fn(logits_classes, gt_classes, fg_index, bg_index)

  loss_bboxes = bboxes_weights * create_loss_bboxes_fn(logits_bboxes, gt_bboxes, fg_index)

  return loss_classes, loss_bboxes


# ------------------------------------------------------------------------
# Detection head
# ------------------------------------------------------------------------ 

def mpi_allreduce_sum(values, *, comm):
    if comm.Get_size() == 1:
        return values
    orig_dtype = values.dtype
    if hvd is None:
        orig_shape = values.shape
        def _allreduce(vals):
            buf = np.zeros(vals.shape, np.float32)
            comm.Allreduce(vals, buf, op=MPI.SUM)
            return buf
        values = tf.py_func(_allreduce, [values], tf.float32)
        values.set_shape(orig_shape)
    else:
        values = hvd.mpi_ops._allreduce(values)
    return tf.cast(values, dtype=orig_dtype) 

def convert_idx_to_token_tensor(inputs, idx2token):
    '''Converts int32 tensor to string tensor.
    inputs: 1d int32 tensor. indices.
    idx2token: dictionary

    Returns
    1d string tensor.
    '''
    def my_func(inputs):
        return " ".join(idx2token[elem] for elem in inputs)

    return tf.py_func(my_func, [inputs], tf.string)

# # def pad(x, maxlen):
# #     '''Pads x, list of sequences, and make it as a numpy array.
# #     x: list of sequences. e.g., [[2, 3, 4], [5, 6, 7, 8, 9], ...]
# #     maxlen: scalar
# #
# #     Returns
# #     numpy int32 array of (len(x), maxlen)
# #     '''
# #     padded = []
# #     for seq in x:
# #         seq += [0] * (maxlen - len(seq))
# #         padded.append(seq)
# #
# #     arry = np.array(padded, np.int32)
# #     assert arry.shape == (len(x), maxlen), "Failed to make an array"
#
#     return arry 

def get_answer_op(context, context_words, answer_start, answer_end):
  return tf.py_func(
      _enum_fn(_get_answer), [context, context_words, answer_start, answer_end],
      'string') 

def sample(pred, placeholders, block_id):
    uni = tf.distributions.Uniform(low=0.0, high=1.0)
    faces = placeholders['faces_triangle'][block_id - 1]
    tilefaces = tf.py_func(choice_faces, [pred, faces], tf.int32)

    num_of_tile_faces = tf.shape(tilefaces)[0]

    xs = tf.gather(pred, tilefaces[:, 0])
    ys = tf.gather(pred, tilefaces[:, 1])
    zs = tf.gather(pred, tilefaces[:, 2])

    u = tf.sqrt(uni.sample([num_of_tile_faces, 1]))
    v = uni.sample([num_of_tile_faces, 1])
    points = (1 - u) * xs + (u * (1 - v)) * ys + u * v * zs
    return points 

def anchor_decoder(boxes, scores, all_anchors, ih, iw, scope='AnchorDecoder'):
  
  with tf.name_scope(scope) as sc:
    final_boxes, classes, scores = \
      tf.py_func(anchor.decode,
                 [boxes, scores, all_anchors, ih, iw],
                 [tf.float32, tf.int32, tf.float32])
    final_boxes = tf.convert_to_tensor(final_boxes, name='boxes')
    classes = tf.convert_to_tensor(tf.cast(classes, tf.int32), name='classes')
    scores = tf.convert_to_tensor(scores, name='scores')
    final_boxes = tf.reshape(final_boxes, (-1, 4))
    classes = tf.reshape(classes, (-1, ))
    scores = tf.reshape(scores, (-1, ))
  
  return final_boxes, classes, scores