Python v1.argmin方法代码示例

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import argmin [as 别名]
def GetDiscreteActionLoss(logits, action_labels, bin_centers, num_bins):
  """Convert labels to one-hot, compute cross-entropy loss, and return loss.

  Args:
    logits: Tensor corresponding to the predicted action logits, with size
      [batch_size, timesteps, action_dim*num_bins]
    action_labels: Tensor corresponding to the real valued action labels, with
      size [batch_size, 1, timesteps, action_dim]
    bin_centers: numpy array of size [num_bins, action_dim] corresponding to
      the centers of each bin for each dimension.
    num_bins: number of discrete bins.
  Returns:
    Scalar tensor, cross entropy loss between the predicted actions and labels.
  """
  action_labels = tf.expand_dims(action_labels, -2)
  bin_centers = tf.constant(bin_centers, dtype=tf.float32)
  while len(bin_centers.shape) < len(action_labels.shape):
    bin_centers = tf.expand_dims(bin_centers, 0)
  discrete_labels = tf.argmin((action_labels - bin_centers)**2, -2)
  onehot_labels = tf.one_hot(discrete_labels, num_bins)
  onehot_labels = tf.reshape(onehot_labels, (-1, num_bins))
  logits = tf.reshape(logits, (-1, num_bins))
  loss = tf.nn.softmax_cross_entropy_with_logits_v2(onehot_labels, logits)
  loss = tf.reduce_mean(loss)
  return loss 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import argmin [as 别名]
def _get_least_likely_class(label, num_classes, logits):
  target_label = tf.argmin(logits, axis=1, output_type=tf.int64)
  # In the off-chance that the least likely class is the true class, the target
  # class is changed to the be the next index.
  return tf.mod(target_label + tf.cast(
      tf.equal(target_label, tf.cast(label, tf.int64)), tf.int64), num_classes) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import argmin [as 别名]
def _find_interval_containing_new_value(x, new_value):
  """Find the index of x (ascending-ordered) after which new_value occurs."""
  new_value_shape = shape_utils.combined_static_and_dynamic_shape(new_value)[0]
  x_shape = shape_utils.combined_static_and_dynamic_shape(x)[0]
  compare = tf.cast(tf.reshape(new_value, shape=(new_value_shape, 1)) >=
                    tf.reshape(x, shape=(1, x_shape)),
                    dtype=tf.int32)
  diff = compare[:, 1:] - compare[:, :-1]
  interval_idx = tf.argmin(diff, axis=1)
  return interval_idx 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import argmin [as 别名]
def test_forward_argminmax():
    for axis in [None, 0, 1, 2]:
        data = np.random.uniform(size=(8, 4, 9)).astype('float32')
        _test_argx(tf.argmax, data=data, axis=axis)
        _test_argx(tf.argmin, data=data, axis=axis)


#######################################################################
# Variable
# -------- 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import argmin [as 别名]
def parse_learning_rate(step, learning_rate):
  """Returns the learning rate as a tensor."""
  if isinstance(learning_rate, float):
    return learning_rate
  # Learning rate schedule of the form:
  # initial_learning_rate[,learning@steps]*. E.g., "1e-3" or
  # "1e-3,1e-4@15000,1e-5@25000". We use eval to allow learning specified as
  # fractions (e.g., 2/255).
  tokens = learning_rate.split(',')
  first_lr = float(eval(tokens[0]))  # pylint: disable=eval-used
  if len(tokens) == 1:
    return tf.constant(first_lr, dtype=tf.float32)
  # Parse steps.
  init_values = [first_lr]
  final_values = []
  init_step = [0]
  final_step = []
  for t in tokens[1:]:
    if '@' in t:
      lr, boundary = t.split('@', 1)
      is_linear = False
    elif 'S' in t:  # Syntactic sugar to indicate a step.
      lr, boundary = t.split('S', 1)
      is_linear = False
    elif 'L' in t:
      lr, boundary = t.split('L', 1)
      is_linear = True
    else:
      raise ValueError('Unknown specification.')
    lr = float(eval(lr))  # pylint: disable=eval-used
    init_values.append(lr)
    if is_linear:
      final_values.append(lr)
    else:
      final_values.append(init_values[-2])
    boundary = int(boundary)
    init_step.append(boundary)
    final_step.append(boundary)
  large_step = max(final_step) + 1
  final_step.append(large_step)
  final_values.append(lr)

  # Find current index.
  boundaries = list(final_step) + [large_step + 2]
  boundaries = tf.convert_to_tensor(boundaries, dtype=tf.int64)
  b = boundaries - tf.minimum(step + 1, large_step + 1)
  large_step = tf.constant(
      large_step, shape=boundaries.shape, dtype=step.dtype)
  b = tf.where(b < 0, large_step, b)
  idx = tf.minimum(tf.argmin(b), len(init_values) - 1)

  init_step = tf.convert_to_tensor(init_step, dtype=tf.float32)
  final_step = tf.convert_to_tensor(final_step, dtype=tf.float32)
  init_values = tf.convert_to_tensor(init_values, dtype=tf.float32)
  final_values = tf.convert_to_tensor(final_values, dtype=tf.float32)
  x1 = tf.gather(init_step, idx)
  x2 = tf.gather(final_step, idx)
  y1 = tf.gather(init_values, idx)
  y2 = tf.gather(final_values, idx)
  return (tf.cast(step, tf.float32) - x1) / (x2 - x1) * (y2 - y1) + y1 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import argmin [as 别名]
def fill_in_the_blank_sized(
    dataset,
    size_bins=(1, 2, 4, 8, 16, 32, 64, 128, 256, 512),
    text_key='text',
    label='fill: '):
  """Fill in the blank preprocessor that labels blank with a binned size.

  The actual blank size is sampled uniformly from the inclusive range of the min
  and max bin. The blank is then filled in with the closest bin size to the
  actual blank size.

  Args:
    dataset: a tf.data.Dataset, the dataset to preprocess.
    size_bins: a list, a list of blank sizes to select from when labelling the
      blank.
    text_key: a string, the key for the text feature to preprocess in the
      dataset examples.
    label: a string, the label to prepend to the inputs.

  Returns:
    a tf.data.Dataset
  """
  bins = sorted(size_bins)

  def my_fn(x):
    """Apply transformation."""
    words = x['words']
    n_words = tf.size(words)

    blank_size = tf.random.uniform(
        [], minval=bins[0], maxval=tf.math.minimum(n_words, bins[-1]),
        dtype=tf.dtypes.int32)
    bin_delta = tf.math.abs(bins - blank_size)
    bin_ = tf.gather(bins, tf.argmin(bin_delta))
    blank_start = tf.random.uniform(
        [], minval=0, maxval=tf.math.maximum(0, n_words-blank_size) + 1,
        dtype=tf.dtypes.int32)

    pre_blank = tf.strings.reduce_join(words[0:blank_start], separator=' ')
    post_blank = tf.strings.reduce_join(
        words[blank_start+blank_size:], separator=' ')
    blank = tf.strings.format('_{}_', bin_)
    # We strip to handle cases where blank is at beginning or end.
    input_ = tf.strings.strip(
        tf.strings.join([pre_blank, blank, post_blank], ' '))
    input_ = tf.strings.join([label, input_])
    target = tf.strings.reduce_join(
        words[blank_start:blank_start+blank_size], separator=' ')
    return {
        'inputs': tf.strings.strip(input_),
        'targets': tf.strings.strip(target)}
  dataset = _split_text_to_words(dataset, text_key, min_num_words=2)
  # Filter out examples with fewer words than the minimum.
  dataset = dataset.filter(lambda x: tf.size(x['words']) >= bins[0])
  dataset = dataset.map(my_fn, num_parallel_calls=tf.data.experimental.AUTOTUNE)
  return dataset