Python tensorflow.compat.v1.square() Examples

def cv_squared(x):
  """The squared coefficient of variation of a sample.

  Useful as a loss to encourage a positive distribution to be more uniform.
  Epsilons added for numerical stability.
  Returns 0 for an empty Tensor.

  Args:
    x: a `Tensor`.

  Returns:
    a `Scalar`.
  """
  epsilon = 1e-10
  float_size = tf.to_float(tf.size(x)) + epsilon
  mean = tf.reduce_sum(x) / float_size
  variance = tf.reduce_sum(tf.squared_difference(x, mean)) / float_size
  return variance / (tf.square(mean) + epsilon) 

def proto_maml_fc_bias(self, prototypes, zero_pad_to_max_way=False):
    """Computes the Prototypical MAML fc layer's bias.

    Args:
      prototypes: Tensor of shape [num_classes, embedding_size]
      zero_pad_to_max_way: Whether to zero padd to max num way.

    Returns:
      fc_bias: Tensor of shape [num_classes] or [self.logit_dim]
        when zero_pad_to_max_way is True.
    """
    fc_bias = -tf.square(tf.norm(prototypes, axis=1))
    if zero_pad_to_max_way:
      paddings = [[0, self.logit_dim - tf.shape(fc_bias)[0]]]
      fc_bias = tf.pad(fc_bias, paddings, 'CONSTANT', constant_values=0)
    return fc_bias 

def _compute_loss(self, prediction_tensor, target_tensor, weights):
    """Compute loss function.

    Args:
      prediction_tensor: A float tensor of shape [batch_size, num_anchors,
        code_size] representing the (encoded) predicted locations of objects.
      target_tensor: A float tensor of shape [batch_size, num_anchors,
        code_size] representing the regression targets
      weights: a float tensor of shape [batch_size, num_anchors]

    Returns:
      loss: a float tensor of shape [batch_size, num_anchors] tensor
        representing the value of the loss function.
    """
    weighted_diff = (prediction_tensor - target_tensor) * tf.expand_dims(
        weights, 2)
    square_diff = 0.5 * tf.square(weighted_diff)
    return tf.reduce_sum(square_diff, 2) 

def normalized_mean_square_error(output, target):
    """Return the TensorFlow expression of normalized mean-squre-error of two distributions.

    Parameters
    ----------
    output : 2D or 4D tensor.
    target : 2D or 4D tensor.
    """
    with tf.name_scope("mean_squared_error_loss"):
        if output.get_shape().ndims == 2:   # [batch_size, n_feature]
            nmse_a = tf.sqrt(tf.reduce_sum(tf.squared_difference(output, target), axis=1))
            nmse_b = tf.sqrt(tf.reduce_sum(tf.square(target), axis=1))
        elif output.get_shape().ndims == 4: # [batch_size, w, h, c]
            nmse_a = tf.sqrt(tf.reduce_sum(tf.squared_difference(output, target), axis=[1,2,3]))
            nmse_b = tf.sqrt(tf.reduce_sum(tf.square(target), axis=[1,2,3]))
        nmse = tf.reduce_mean(nmse_a / nmse_b)
    return nmse 

def setUp(self):
        super(PersistentOpEvaluatorTest, self).setUp()

        patch = tf.test.mock.patch(
            "tensorflow.compat.v1.Session", wraps=tf.Session
        )
        patch.start()
        self.addCleanup(patch.stop)

        class Squarer(op_evaluator.PersistentOpEvaluator):
            def __init__(self):
                super(Squarer, self).__init__()
                self._input = None
                self._squarer = None

            def initialize_graph(self):
                self._input = tf.placeholder(tf.int32)
                self._squarer = tf.square(self._input)

            def run(self, xs):  # pylint: disable=arguments-differ
                return self._squarer.eval(feed_dict={self._input: xs})

        self._square = Squarer() 

def _get_cost_function(self):
        """Compute the cost of the Mittens objective function.

        If self.mittens = 0, this is the same as the cost of GloVe.
        """
        self.weights = tf.placeholder(
            tf.float32, shape=[self.n_words, self.n_words])
        self.log_coincidence = tf.placeholder(
            tf.float32, shape=[self.n_words, self.n_words])
        self.diffs = tf.subtract(self.model, self.log_coincidence)
        cost = tf.reduce_sum(
            0.5 * tf.multiply(self.weights, tf.square(self.diffs)))
        if self.mittens > 0:
            self.mittens = tf.constant(self.mittens, tf.float32)
            cost += self.mittens * tf.reduce_sum(
                tf.multiply(
                    self.has_embedding,
                    self._tf_squared_euclidean(
                        tf.add(self.W, self.C),
                        self.original_embedding)))
        tf.summary.scalar("cost", cost)
        return cost 

def bottleneck(self, x):
    z_size = self.hparams.bottleneck_bits
    x_shape = common_layers.shape_list(x)
    with tf.variable_scope('bottleneck', reuse=tf.AUTO_REUSE):
      mu = x[..., :self.hparams.bottleneck_bits]
      if self.hparams.mode != tf.estimator.ModeKeys.TRAIN:
        return mu, 0.0  # No sampling or kl loss on eval.
      log_sigma = x[..., self.hparams.bottleneck_bits:]
      epsilon = tf.random_normal(x_shape[:-1] + [z_size])
      z = mu + tf.exp(log_sigma / 2) * epsilon
      kl = 0.5 * tf.reduce_mean(
          tf.exp(log_sigma) + tf.square(mu) - 1. - log_sigma, axis=-1)
      # This is the 'free bits' trick mentioned in Kingma et al. (2016)
      free_bits = self.hparams.free_bits
      kl_loss = tf.reduce_mean(tf.maximum(kl - free_bits, 0.0))
    return z, kl_loss * self.hparams.kl_beta 

def lazy_square(tensor):
  """Computes the square of a tensor in a lazy way.

  This function is lazy in the following sense, for:
    tensor = tf.sqrt(input)
  will return input (and not tf.square(tensor)).

  Args:
    tensor: A `Tensor` of floats to compute the square of.

  Returns:
    The square of the input tensor.
  """
  if tensor.op.type == 'Sqrt':
    return tensor.op.inputs[0]
  else:
    return tf.square(tensor) 

def test_group_lasso_conv3d(self):
    shape = [3, 3, 3]
    video = tf.zeros([2, 3, 3, 3, 1])
    net = slim.conv3d(
        video,
        5,
        shape,
        padding='VALID',
        weights_initializer=tf.glorot_normal_initializer(),
        scope='vconv1')
    conv3d_op = tf.get_default_graph().get_operation_by_name('vconv1/Conv3D')
    conv3d_weights = conv3d_op.inputs[1]

    threshold = 0.09
    flop_reg = flop_regularizer.GroupLassoFlopsRegularizer([net.op],
                                                           threshold=threshold)
    norm = tf.sqrt(tf.reduce_mean(tf.square(conv3d_weights), [0, 1, 2, 3]))
    alive = tf.reduce_sum(tf.cast(norm > threshold, tf.float32))
    with self.session():
      flop_coeff = 2 * shape[0] * shape[1] * shape[2]
      tf.compat.v1.global_variables_initializer().run()
      self.assertAllClose(flop_reg.get_cost(), flop_coeff * alive)
      self.assertAllClose(flop_reg.get_regularization_term(),
                          flop_coeff * tf.reduce_sum(norm)) 

def vae(x, z_size, name=None):
  """Simple variational autoencoder without discretization.

  Args:
    x: Input to the discretization bottleneck.
    z_size: Number of bits, where discrete codes range from 1 to 2**z_size.
    name: Name for the bottleneck scope.

  Returns:
    Embedding function, latent, loss, mu and log_simga.
  """
  with tf.variable_scope(name, default_name="vae"):
    mu = tf.layers.dense(x, z_size, name="mu")
    log_sigma = tf.layers.dense(x, z_size, name="log_sigma")
    shape = common_layers.shape_list(x)
    epsilon = tf.random_normal([shape[0], shape[1], 1, z_size])
    z = mu + tf.exp(log_sigma / 2) * epsilon
    kl = 0.5 * tf.reduce_mean(
        tf.expm1(log_sigma) + tf.square(mu) - log_sigma, axis=-1)
    free_bits = z_size // 4
    kl_loss = tf.reduce_mean(tf.maximum(kl - free_bits, 0.0))
    return z, kl_loss, mu, log_sigma 

def bottleneck(self, x):
    hparams = self.hparams
    z_size = hparams.bottleneck_bits
    x_shape = common_layers.shape_list(x)
    with tf.variable_scope("vae"):
      mu = tf.layers.dense(x, z_size, name="mu")
      if hparams.mode != tf.estimator.ModeKeys.TRAIN:
        return mu, 0.0  # No sampling or kl loss on eval.
      log_sigma = tf.layers.dense(x, z_size, name="log_sigma")
      epsilon = tf.random_normal(x_shape[:-1] + [z_size])
      z = mu + tf.exp(log_sigma / 2) * epsilon
      kl = 0.5 * tf.reduce_mean(
          tf.expm1(log_sigma) + tf.square(mu) - log_sigma, axis=-1)
      free_bits = z_size // 4
      kl_loss = tf.reduce_mean(tf.maximum(kl - free_bits, 0.0))
    return z, kl_loss * hparams.kl_beta 

def _test_l2_normalize(ishape, eps, axis):
    """ testing l2 normalize (uses max, sum, square, sqrt frontend operators)"""

    inp_array = np.random.uniform(size=ishape).astype(np.float32)

    with tf.Graph().as_default():
        in1 = tf.placeholder(shape=inp_array.shape, dtype=inp_array.dtype)
        nn.l2_normalize(in1,
                        axis=axis,
                        epsilon=eps,
                        name=None,
                        dim=None)

        compare_tf_with_tvm(inp_array, 'Placeholder:0', 'l2_normalize:0') 

def variable_summaries(var, scope=""):
  """Attach a lot of summaries to a Tensor (for TensorBoard visualization)."""
  with tf.name_scope(scope):
    with tf.name_scope("summaries"):
      mean = tf.reduce_mean(var)
      tf.summary.scalar("mean", mean)
      with tf.name_scope("stddev"):
        stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))
      tf.summary.scalar("stddev", stddev)
      tf.summary.scalar("max", tf.reduce_max(var))
      tf.summary.scalar("min", tf.reduce_min(var))
      tf.summary.histogram("histogram", var) 

def _grad_variance(self):
    """Estimate of gradient Variance.

    Returns:
      C_t ops.
    """
    grad_var_ops = []
    tensor_to_avg = []
    for t, g in zip(self._vars, self._grad):
      if isinstance(g, tf.IndexedSlices):
        tensor_to_avg.append(
            tf.reshape(tf.unsorted_segment_sum(g.values,
                                               g.indices,
                                               g.dense_shape[0]),
                       shape=t.get_shape()))
      else:
        tensor_to_avg.append(g)
    avg_op = self._moving_averager.apply(tensor_to_avg)
    grad_var_ops.append(avg_op)
    with tf.control_dependencies([avg_op]):
      self._grad_avg = [self._moving_averager.average(val)
                        for val in tensor_to_avg]
      self._grad_avg_squared = [tf.square(val) for val in self._grad_avg]

    # Compute Variance
    self._grad_var = tf.maximum(
        tf.constant(1e-6, dtype=self._grad_norm_squared_avg.dtype),
        self._grad_norm_squared_avg
        - tf.add_n([tf.reduce_sum(val) for val in self._grad_avg_squared]))
    if self._sparsity_debias:
      self._grad_var *= self._sparsity_avg
    return grad_var_ops  # C_t 

def td_learning(v_tm1, r_t, pcont_t, v_t, name="TDLearning"):
  """Implements the TD(0)-learning loss as a TensorFlow op.

  The TD loss is `0.5` times the squared difference between `v_tm1` and
  the target `r_t + pcont_t * v_t`.

  See "Learning to Predict by the Methods of Temporal Differences" by Sutton.
  (https://link.springer.com/article/10.1023/A:1022633531479).

  Args:
    v_tm1: Tensor holding values at previous timestep, shape `[B]`.
    r_t: Tensor holding rewards, shape `[B]`.
    pcont_t: Tensor holding pcontinue values, shape `[B]`.
    v_t: Tensor holding values at current timestep, shape `[B]`.
    name: name to prefix ops created by this function.

  Returns:
    A namedtuple with fields:

    * `loss`: a tensor containing the batch of losses, shape `[B]`.
    * `extra`: a namedtuple with fields:
        * `target`: batch of target values for `v_tm1`, shape `[B]`.
        * `td_error`: batch of temporal difference errors, shape `[B]`.
  """
  # Rank and compatibility checks.
  base_ops.wrap_rank_shape_assert([[v_tm1, v_t, r_t, pcont_t]], [1], name)

  # TD(0)-learning op.
  with tf.name_scope(name, values=[v_tm1, r_t, pcont_t, v_t]):

    # Build target.
    target = tf.stop_gradient(r_t + pcont_t * v_t)

    # Temporal difference error and loss.
    # Loss is MSE scaled by 0.5, so the gradient is equal to the TD error.
    td_error = target - v_tm1
    loss = 0.5 * tf.square(td_error)
    return base_ops.LossOutput(loss, TDExtra(target, td_error)) 

def qv_max(v_tm1, r_t, pcont_t, q_t, name="QVMAX"):
  """Implements the QVMAX learning loss as a TensorFlow op.

  The QVMAX loss is `0.5` times the squared difference between `v_tm1` and
  the target `r_t + pcont_t * max q_t`, where `q_t` is separately learned
  through QV learning (c.f. `action_value_ops.qv_learning`).

  See "The QV Family Compared to Other Reinforcement Learning Algorithms" by
  Wiering and van Hasselt (2009).
  (http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.713.1931)

  Args:
    v_tm1: Tensor holding values at previous timestep, shape `[B]`.
    r_t: Tensor holding rewards, shape `[B]`.
    pcont_t: Tensor holding pcontinue values, shape `[B]`.
    q_t: Tensor of action values at current timestep, shape `[B, num_actions]`.
    name: name to prefix ops created by this function.

  Returns:
    A namedtuple with fields:

    * `loss`: a tensor containing the batch of losses, shape `[B]`.
    * `extra`: a namedtuple with fields:
        * `target`: batch of target values for `v_tm1`, shape `[B]`.
        * `td_error`: batch of temporal difference errors, shape `[B]`.
  """
  # Rank and compatibility checks.
  base_ops.wrap_rank_shape_assert([[v_tm1, r_t, pcont_t], [q_t]], [1, 2], name)

  # The QVMAX op.
  with tf.name_scope(name, values=[v_tm1, r_t, pcont_t, q_t]):

    # Build target.
    target = tf.stop_gradient(r_t + pcont_t * tf.reduce_max(q_t, axis=1))

    # Temporal difference error and loss.
    # Loss is MSE scaled by 0.5, so the gradient is equal to the TD error.
    td_error = target - v_tm1
    loss = 0.5 * tf.square(td_error)
    return base_ops.LossOutput(loss, TDExtra(target, td_error)) 

def variable_summaries(var):
  """Attach a lot of summaries to a Tensor (for TensorBoard visualization)."""
  with tf.name_scope('summaries'):
    mean = tf.reduce_mean(var)
    tf.summary.scalar('mean', mean)
    with tf.name_scope('stddev'):
      stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))
    tf.summary.scalar('stddev', stddev)
    tf.summary.scalar('max', tf.reduce_max(var))
    tf.summary.scalar('min', tf.reduce_min(var))
    tf.summary.histogram('histogram', var) 

def _test_l2_pool2d(input_shape, ksize, strides, padding, data_format, fused_func_name=None):
    x = np.arange(np.prod(input_shape), dtype=np.float32).reshape(input_shape) - 1

    with tf.Graph().as_default():
        in_data = tf.placeholder(
            dtype=tf.float32, name="input", shape=input_shape)
        out = tf.sqrt(tf.nn.avg_pool(
            tf.square(in_data), ksize=ksize, strides=strides,
            padding=padding, data_format=data_format))
        out = with_fused_activation_function(out, fused_func_name)

        compare_tflite_with_tvm(x, 'input', [in_data], [out]) 

def _test_square(data):
    """ One iteration of square """
    return _test_unary_elemwise(math_ops.square, data)

#######################################################################
# Elu
# --- 

def vq_nearest_neighbor(x, means,
                        soft_em=False, num_samples=10, temperature=None):
  """Find the nearest element in means to elements in x."""
  bottleneck_size = common_layers.shape_list(means)[0]
  x_norm_sq = tf.reduce_sum(tf.square(x), axis=-1, keepdims=True)
  means_norm_sq = tf.reduce_sum(tf.square(means), axis=-1, keepdims=True)
  scalar_prod = tf.matmul(x, means, transpose_b=True)
  dist = x_norm_sq + tf.transpose(means_norm_sq) - 2 * scalar_prod
  if soft_em:
    x_means_idx = tf.multinomial(-dist, num_samples=num_samples)
    x_means_hot = tf.one_hot(
        x_means_idx, depth=common_layers.shape_list(means)[0])
    x_means_hot = tf.reduce_mean(x_means_hot, axis=1)
  else:
    if temperature is None:
      x_means_idx = tf.argmax(-dist, axis=-1)
    else:
      x_means_idx = tf.multinomial(- dist / temperature, 1)
      x_means_idx = tf.squeeze(x_means_idx, axis=-1)
    if (common_layers.should_generate_summaries() and
        not common_layers.is_xla_compiled()):
      tf.summary.histogram("means_idx", tf.reshape(x_means_idx, [-1]))
    x_means_hot = tf.one_hot(x_means_idx, bottleneck_size)
  x_means_hot_flat = tf.reshape(x_means_hot, [-1, bottleneck_size])
  x_means = tf.matmul(x_means_hot_flat, means)
  e_loss = tf.reduce_mean(tf.squared_difference(x, tf.stop_gradient(x_means)))
  return x_means_hot, e_loss, dist 

def test_forward_square():
    """test operator Square """
    np_data = np.random.uniform(1, 100, size=(2, 3, 5)).astype(np.float32)
    tf.reset_default_graph()
    with tf.Graph().as_default():
        in_data = tf.placeholder(tf.float32, (2, 3, 5), name="in_data")
        tf.square(in_data, name="square")
        compare_tf_with_tvm([np_data], ['in_data:0'], 'square:0') 

def test_multiple_cond_vars():
    graph = tf.Graph()
    with graph.as_default():
        x1 = tf.constant(7)
        x2 = tf.constant(12)
        z = tf.constant(20)
        r = tf.cond(tf.less(tf.add(x1, x2), 10),
                    lambda: tf.add(10, 2), lambda: tf.square(5))

        with tf.Session() as sess:
            tf_out = sess.run(r)

    check_equal(graph, tf_out) 

def test_vanilla_loop_bound():
    graph = tf.Graph()
    with graph.as_default():
        dshape = (2, 10)
        dtype = "float32"
        dname = "data"
        np_data = np.random.uniform(size=dshape).astype(dtype)
        data = tf.placeholder(shape=dshape, dtype=dtype, name=dname)
        x = tf.slice(data, [1, 4], [1, 4])
        outer = x + 5.0
        def body(x, y):
            res = tf.cond(tf.less(y, 10), lambda: tf.add(
                10.0, 20.0), lambda: tf.square(10.0))
            z = tf.constant(7)
            res = tf.cond(tf.less(z, 10), lambda: res * 5, lambda: res + 10)
            return tf.multiply(res, x * outer), y + 1

        y = tf.constant(0)
        def condition(x, y):
            return tf.less(y, 20)

        r = tf.while_loop(condition, body, loop_vars=[x, y])
        with tf.Session() as sess:
            tf_out = sess.run(r, feed_dict={"%s:0" % dname: np_data})

    check_equal(graph, tf_out, {dname: np_data}) 

def test_nested_loop_bound():
    graph = tf.Graph()
    with graph.as_default():
        dshape = (2, 10)
        dtype = "float32"
        dname = "data"
        np_data = np.random.uniform(size=dshape).astype(dtype)
        data = tf.placeholder(shape=dshape, dtype=dtype, name=dname)
        x = tf.slice(data, [1, 4], [1, 4])
        outer = x + 5.0
        def body(x, y):
            res = tf.cond(tf.less(y, 10), lambda: tf.add(
                10.0, 20.0), lambda: tf.square(10.0))
            def nested_body(nx, ny):
                return nx + 1, res + 2.0
            def nested_cond(nx, ny):
                return tf.less(nx, 15)
            nx = tf.constant(0)
            ny = tf.constant(0.0)
            nested_res = tf.while_loop(nested_cond, nested_body, loop_vars=[nx, ny])
            res = res + nested_res[1]
            z = tf.constant(7)
            res = tf.cond(tf.less(z, 10), lambda: res * 5, lambda: res + 10)
            return tf.multiply(res, x * outer), y + 1

        y = tf.constant(0)
        def condition(x, y):
            return tf.less(y, 20)

        r = tf.while_loop(condition, body, loop_vars=[x, y])
        with tf.Session() as sess:
            tf_out = sess.run(r, feed_dict={"%s:0" % dname: np_data})

    check_equal(graph, tf_out, {dname: np_data}) 

def compute_logits(self, support_embeddings, query_embeddings,
                     onehot_support_labels):
    """Computes the negative distances of each query point to each prototype."""

    # [num test images, 1, embedding size].
    query_embeddings = tf.expand_dims(query_embeddings, 1)

    prototypes = compute_prototypes(support_embeddings, onehot_support_labels)

    # [1, num_clases, embedding_size].
    prototypes = tf.expand_dims(prototypes, 0)

    # Squared euclidean distances between each test embedding / prototype pair.
    distances = tf.reduce_sum(tf.square(query_embeddings - prototypes), 2)
    return -distances 

def _make_add_squared_grads(self):
    assignments = []
    for sum_squared_grads, grads in zip(self._sum_squared_grads, self._grads):
      assignments.append(sum_squared_grads.assign_add(tf.square(grads)))
    return tf.group(assignments + [self._num_squared_grads.assign_add(1)]) 

def _norm(self, x):
    """Compute the safe norm."""
    return tf.sqrt(tf.reduce_sum(tf.square(x), keepdims=True, axis=-1) + 1e-7) 

def _compute_sample_loss(self, gt, output):
    """Compute point sampling loss."""
    sample_loss = tf.square(gt - output)
    if self._sample_bbx > 0:
      loss_bbx = sample_loss[:, :self._sample_bbx]
      loss_bbx = tf.reduce_mean(loss_bbx)
    else:
      loss_bbx = 0.
    if self._sample_surf > 0:
      loss_surf = sample_loss[:, self._sample_bbx:]
      loss_surf = tf.reduce_mean(loss_surf)
    else:
      loss_surf = 0.
    sample_loss = loss_bbx + 0.1 * loss_surf
    return sample_loss 

def _compute_bbx_loss(self, trans, pts, gt):
    """Compute bounding box loss."""
    oo = 1e5
    inside = tf.expand_dims(tf.cast(gt > 0.5, tf.float32), axis=1)
    trans = tf.expand_dims(trans, axis=2)
    pts = tf.expand_dims(pts, axis=1)
    distances = tf.reduce_sum(tf.square(trans - pts), axis=-1, keepdims=True)
    distances = inside * distances + (1 - inside) * oo
    min_dis = tf.reduce_min(distances, axis=2)
    return tf.reduce_mean(min_dis) 

def _compute_overlap_loss(self, overlap):
    """Compute overlap loss."""
    return tf.reduce_mean(tf.square(tf.nn.relu(overlap - 2.)))