def sparse_softmax_cross_entropy(logits, labels, weights=1.0, scope=None):
  """Cross-entropy loss using `tf.nn.sparse_softmax_cross_entropy_with_logits`.

  `weights` acts as a coefficient for the loss. If a scalar is provided,
  then the loss is simply scaled by the given value. If `weights` is a
  tensor of size [`batch_size`], then the loss weights apply to each
  corresponding sample.

  Args:
    logits: [batch_size, num_classes] logits outputs of the network .
    labels: [batch_size, 1] or [batch_size] labels of dtype `int32` or `int64`
      in the range `[0, num_classes)`.
    weights: Coefficients for the loss. The tensor must be a scalar or a tensor
      of shape [batch_size] or [batch_size, 1].
    scope: the scope for the operations performed in computing the loss.

  Returns:
    A scalar `Tensor` representing the mean loss value.

  Raises:
    ValueError: If the shapes of `logits`, `labels`, and `weights` are
      incompatible, or if `weights` is None.
  """
  with ops.name_scope(scope, "sparse_softmax_cross_entropy_loss",
                      [logits, labels, weights]) as scope:
    labels = array_ops.reshape(labels, shape=[array_ops.shape(labels)[0]])

    losses = nn.sparse_softmax_cross_entropy_with_logits(labels=labels,
                                                         logits=logits,
                                                         name="xentropy")
    return compute_weighted_loss(losses, weights, scope=scope)

def _softmax_cross_entropy_loss(logits, target):
  check_shape_op = control_flow_ops.Assert(
      math_ops.less_equal(array_ops.rank(target), 2),
      ["target's shape should be either [batch_size, 1] or [batch_size]"])
  with ops.control_dependencies([check_shape_op]):
    target = array_ops.reshape(target, shape=[array_ops.shape(target)[0]])
  return nn.sparse_softmax_cross_entropy_with_logits(logits, target)

def _softmax_cross_entropy_loss(logits, labels):
  # Check that we got integer for classification.
  if not labels.dtype.is_integer:
    raise ValueError("Labels dtype should be integer "
                     "Instead got %s." % labels.dtype)
  # sparse_softmax_cross_entropy_with_logits requires [batch_size] labels.
  if len(labels.get_shape()) == 2:
    labels = array_ops.squeeze(labels, squeeze_dims=[1])
  return nn.sparse_softmax_cross_entropy_with_logits(logits, labels)

def model():
    print("building model ...")
    with tf.variable_scope('train'):
        print("building model ...")
        X_pl = tf.placeholder(tf.float32, [None, num_features])
        X_expand = tf.expand_dims(X_pl, axis=2)
        print("X_pl", X_pl.get_shape())
        t_pl = tf.placeholder(tf.int32, [None,])
        print("t_pl", t_pl.get_shape())
        is_training_pl = tf.placeholder(tf.bool)
        cell_fw = tf.nn.rnn_cell.GRUCell(205)
        cell_bw = tf.nn.rnn_cell.GRUCell(205)
        seq_len = tf.reduce_sum(tf.ones(tf.shape(X_pl), dtype=tf.int32), axis=1)
        _, enc_states = tf.nn.bidirectional_dynamic_rnn(cell_fw=cell_fw,
            cell_bw=cell_bw, inputs=X_expand, sequence_length=seq_len,
            dtype=tf.float32)
        enc_states = tf.concat(1, enc_states)
        enc_states_drop = dropout(enc_states, is_training=is_training_pl) 
        l1 = fully_connected(enc_states_drop, 200, activation_fn=None)
        l1 = batch_norm(l1, is_training=is_training_pl)
        l1_relu = relu(l1)
        l1_dropout = dropout(l1_relu, is_training=is_training_pl)
        l2 = fully_connected(l1_dropout, 200, activation_fn=None)
        l2 = batch_norm(l2, is_training=is_training_pl)
        l2_relu = relu(l2)
        l_out = fully_connected(l2_relu, num_outputs=num_classes, activation_fn=None)
        l_out_softmax = tf.nn.softmax(l_out)
        tf.contrib.layers.summarize_variables()

    with tf.variable_scope('metrics'):
        loss = sparse_softmax_cross_entropy_with_logits(l_out, t_pl)
        print("loss", loss.get_shape())
        loss = tf.reduce_mean(loss)
        print("loss", loss.get_shape())
        tf.summary.scalar('train/loss', loss)
        argmax = tf.to_int32(tf.argmax(l_out, 1))
        print("argmax", argmax.get_shape())
        correct = tf.to_float(tf.equal(argmax, t_pl))
        print("correct,", correct.get_shape())
        accuracy = tf.reduce_mean(correct)
        print("accuracy", accuracy.get_shape())

    with tf.variable_scope('optimizer'):
        print("building optimizer ...")
        global_step = tf.Variable(0, name='global_step', trainable=False)
        optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
        grads_and_vars = optimizer.compute_gradients(loss)
        gradients, variables = zip(*grads_and_vars)
        clipped_gradients, global_norm = (
            tf.clip_by_global_norm(gradients, clip_norm))
        clipped_grads_and_vars = zip(clipped_gradients, variables)

        tf.summary.scalar('train/global_gradient_norm', global_norm)

        train_op = optimizer.apply_gradients(clipped_grads_and_vars, global_step=global_step)

    return X_pl, t_pl, is_training_pl, l_out, l_out_softmax, loss, accuracy, train_op, global_step

def _softmax_cross_entropy_loss(logits, target):
  # Check that we got integer for classification.
  if not target.dtype.is_integer:
    raise ValueError("Target's dtype should be integer "
                     "Instead got %s." % target.dtype)
  # sparse_softmax_cross_entropy_with_logits requires [batch_size] target.
  if len(target.get_shape()) == 2:
    target = array_ops.squeeze(target, squeeze_dims=[1])
  loss_vec = nn.sparse_softmax_cross_entropy_with_logits(logits, target)
  return loss_vec

def _softmax_cross_entropy_loss(logits, target):
    # sigmoid_cross_entropy_with_logits requires [batch_size, 1] target.
    # Check that we got int32/int64 for classification.
    if not target.dtype.is_compatible_with(dtypes.int64) and not target.dtype.is_compatible_with(dtypes.int32):
        raise ValueError("Target's dtype should be int32, int64 or compatible. " "Instead got %s." % target.dtype)
    # sparse_softmax_cross_entropy_with_logits requires [batch_size] target.
    if len(target.get_shape()) == 2:
        target = array_ops.squeeze(target, squeeze_dims=[1])
    loss_vec = nn.sparse_softmax_cross_entropy_with_logits(logits, target)
    return loss_vec

def sparse_softmax_cross_entropy(
    labels, logits, weights=1.0, scope=None,
    loss_collection=ops.GraphKeys.LOSSES,
    reduction=Reduction.SUM_BY_NONZERO_WEIGHTS):
  """Cross-entropy loss using `tf.nn.sparse_softmax_cross_entropy_with_logits`.

  `weights` acts as a coefficient for the loss. If a scalar is provided,
  then the loss is simply scaled by the given value. If `weights` is a
  tensor of shape `[batch_size]`, then the loss weights apply to each
  corresponding sample.

  Args:
    labels: `Tensor` of shape `[d_0, d_1, ..., d_{r-1}]` (where `r` is rank of
      `labels` and result) and dtype `int32` or `int64`. Each entry in `labels`
      must be an index in `[0, num_classes)`. Other values will raise an
      exception when this op is run on CPU, and return `NaN` for corresponding
      loss and gradient rows on GPU.
    logits: Unscaled log probabilities of shape
      `[d_0, d_1, ..., d_{r-1}, num_classes]` and dtype `float16`, `float32` or
      `float64`.
    weights: Coefficients for the loss. This must be scalar or broadcastable to
      `labels` (i.e. same rank and each dimension is either 1 or the same).
    scope: the scope for the operations performed in computing the loss.
    loss_collection: collection to which the loss will be added.
    reduction: Type of reduction to apply to loss.

  Returns:
    Weighted loss `Tensor` of the same type as `logits`. If `reduction` is
    `NONE`, this has the same shape as `labels`; otherwise, it is scalar.

  Raises:
    ValueError: If the shapes of `logits`, `labels`, and `weights` are
      incompatible, or if any of them are None.

  @compatibility(eager)
  The `loss_collection` argument is ignored when executing eagerly. Consider
  holding on to the return value or collecting losses via a `tf.keras.Model`.
  @end_compatibility
  """
  if labels is None:
    raise ValueError("labels must not be None.")
  if logits is None:
    raise ValueError("logits must not be None.")
  with ops.name_scope(scope, "sparse_softmax_cross_entropy_loss",
                      (logits, labels, weights)) as scope:
    # As documented above in Args, labels contain class IDs and logits contains
    # 1 probability per class ID, so we expect rank(logits) - rank(labels) == 1;
    # therefore, expected_rank_diff=1.
    labels, logits, weights = _remove_squeezable_dimensions(
        labels, logits, weights, expected_rank_diff=1)
    losses = nn.sparse_softmax_cross_entropy_with_logits(labels=labels,
                                                         logits=logits,
                                                         name="xentropy")
    return compute_weighted_loss(
        losses, weights, scope, loss_collection, reduction=reduction)

def model():
    tf.set_random_seed(1)
    print("building model ...")
    with tf.variable_scope('train'):
        print("building model ...")
        X_pl = tf.placeholder(tf.float32, [None, num_features])
        print("X_pl", X_pl.get_shape())
        t_pl = tf.placeholder(tf.int32, [None,])
        print("t_pl", t_pl.get_shape())
        is_training_pl = tf.placeholder(tf.bool)
        X_bn = batch_norm(X_pl, is_training=is_training_pl)
        print("X_bn", X_bn.get_shape())
        l1 = fully_connected(X_pl, num_outputs=100, activation_fn=relu)#, normalizer_fn=batch_norm)
        print("l1", l1.get_shape())
        l1_drop = dropout(l1, is_training=is_training_pl)
        print("l1_drop", l1_drop.get_shape())
        l_out = fully_connected(l1_drop, num_outputs=num_classes, activation_fn=None)
        print("l_out", l_out.get_shape())
        l_out_softmax = tf.nn.softmax(l_out)
        tf.contrib.layers.summarize_variables()

    with tf.variable_scope('metrics'):
        loss = sparse_softmax_cross_entropy_with_logits(l_out, t_pl)
        print("loss", loss.get_shape())
        loss = tf.reduce_mean(loss)
        print("loss", loss.get_shape())
        tf.summary.scalar('train/loss', loss)
        argmax = tf.to_int32(tf.argmax(l_out, 1))
        print("argmax", argmax.get_shape())
        correct = tf.to_float(tf.equal(argmax, t_pl))
        print("correct,", correct.get_shape())
        accuracy = tf.reduce_mean(correct)
        print("accuracy", accuracy.get_shape())

    with tf.variable_scope('optimizer'):
        print("building optimizer ...")
        global_step = tf.Variable(0, name='global_step', trainable=False)
        optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
        grads_and_vars = optimizer.compute_gradients(loss)
        gradients, variables = zip(*grads_and_vars)
        clipped_gradients, global_norm = (
            tf.clip_by_global_norm(gradients, clip_norm))
        clipped_grads_and_vars = zip(clipped_gradients, variables)

        tf.summary.scalar('train/global_gradient_norm', global_norm)

        train_op = optimizer.apply_gradients(clipped_grads_and_vars, global_step=global_step)

    return X_pl, t_pl, is_training_pl, l_out, l_out_softmax, loss, accuracy, train_op, global_step

def deprecated_flipped_sparse_softmax_cross_entropy_with_logits(logits,
                                                                labels,
                                                                name=None):
  """Computes sparse softmax cross entropy between `logits` and `labels`.

  This function diffs from tf.nn.sparse_softmax_cross_entropy_with_logits only
  in the argument order.

  Measures the probability error in discrete classification tasks in which the
  classes are mutually exclusive (each entry is in exactly one class).  For
  example, each CIFAR-10 image is labeled with one and only one label: an image
  can be a dog or a truck, but not both.

  **NOTE:**  For this operation, the probability of a given label is considered
  exclusive.  That is, soft classes are not allowed, and the `labels` vector
  must provide a single specific index for the true class for each row of
  `logits` (each minibatch entry).  For soft softmax classification with
  a probability distribution for each entry, see
  `softmax_cross_entropy_with_logits`.

  **WARNING:** This op expects unscaled logits, since it performs a softmax
  on `logits` internally for efficiency.  Do not call this op with the
  output of `softmax`, as it will produce incorrect results.

  A common use case is to have logits of shape `[batch_size, num_classes]` and
  labels of shape `[batch_size]`. But higher dimensions are supported.

  Args:

    logits: Unscaled log probabilities of rank `r` and shape
      `[d_0, d_1, ..., d_{r-2}, num_classes]` and dtype `float32` or `float64`.
    labels: `Tensor` of shape `[d_0, d_1, ..., d_{r-2}]` and dtype `int32` or
      `int64`. Each entry in `labels` must be an index in `[0, num_classes)`.
      Other values will raise an exception when this op is run on CPU, and
      return `NaN` for corresponding corresponding loss and gradient rows
      on GPU.
    name: A name for the operation (optional).

  Returns:
    A `Tensor` of the same shape as `labels` and of the same type as `logits`
    with the softmax cross entropy loss.

  Raises:
    ValueError: If logits are scalars (need to have rank >= 1) or if the rank
      of the labels is not equal to the rank of the logits minus one.
  """
  return nn.sparse_softmax_cross_entropy_with_logits(
      labels=labels, logits=logits, name=name)

    def _loss(self, logits, target, weight_tensor):
        if self._n_classes < 2:
            loss_vec = math_ops.square(logits - math_ops.to_float(target))
        elif self._n_classes == 2:
            loss_vec = nn.sigmoid_cross_entropy_with_logits(logits, math_ops.to_float(target))
        else:
            loss_vec = nn.sparse_softmax_cross_entropy_with_logits(logits, array_ops.reshape(target, [-1]))

        if weight_tensor is None:
            return math_ops.reduce_mean(loss_vec, name="loss")
        else:
            loss_vec = array_ops.reshape(loss_vec, shape=(-1,))
            loss_vec = math_ops.mul(loss_vec, array_ops.reshape(weight_tensor, shape=(-1,)))
            return math_ops.div(
                math_ops.reduce_sum(loss_vec), math_ops.to_float(math_ops.reduce_sum(weight_tensor)), name="loss"
            )

def sparse_softmax_cross_entropy(
    labels, logits, weights=1.0, scope=None,
    loss_collection=ops.GraphKeys.LOSSES,
    reduction=Reduction.WEIGHTED_SUM_BY_NONZERO_WEIGHTS):
  """Cross-entropy loss using `tf.nn.sparse_softmax_cross_entropy_with_logits`.

  `weights` acts as a coefficient for the loss. If a scalar is provided,
  then the loss is simply scaled by the given value. If `weights` is a
  tensor of shape [`batch_size`], then the loss weights apply to each
  corresponding sample.

  Args:
    labels: `Tensor` of shape `[d_0, d_1, ..., d_{r-1}]` (where `r` is rank of
      `labels` and result) and dtype `int32` or `int64`. Each entry in `labels`
      must be an index in `[0, num_classes)`. Other values will raise an
      exception when this op is run on CPU, and return `NaN` for corresponding
      loss and gradient rows on GPU.
    logits: Unscaled log probabilities of shape
      `[d_0, d_1, ..., d_{r-1}, num_classes]` and dtype `float32` or `float64`.
    weights: Coefficients for the loss. This must be scalar or of same rank as
      `labels`
    scope: the scope for the operations performed in computing the loss.
    loss_collection: collection to which the loss will be added.
    reduction: Type of reduction to apply to loss.

  Returns:
    A scalar `Tensor` that returns the weighted loss.

  Raises:
    ValueError: If the shapes of logits, labels, and weight are incompatible, or
      if `weights` is None.
  """
  with ops.name_scope(scope, "sparse_softmax_cross_entropy_loss",
                      (logits, labels, weights)) as scope:
    # As documented above in Args, labels contain class IDs and logits contains
    # 1 probability per class ID, so we expect rank(logits) - rank(labels) == 1;
    # therefore, expected_rank_diff=1.
    labels, logits, weights = _remove_squeezable_dimensions(
        labels, logits, weights, expected_rank_diff=1)
    losses = nn.sparse_softmax_cross_entropy_with_logits(labels=labels,
                                                         logits=logits,
                                                         name="xentropy")
    return compute_weighted_loss(
        losses, weights, scope, loss_collection, reduction=reduction)

 def _loss_vec(self, logits, target):
   if self._n_classes == 2:
     # sigmoid_cross_entropy_with_logits requires [batch_size, 1] target.
     if len(target.get_shape()) == 1:
       target = array_ops.expand_dims(target, dim=[1])
     loss_vec = nn.sigmoid_cross_entropy_with_logits(
         logits, math_ops.to_float(target))
   else:
     # Check that we got int32/int64 for classification.
     if (not target.dtype.is_compatible_with(dtypes.int64) and
         not target.dtype.is_compatible_with(dtypes.int32)):
       raise ValueError("Target's dtype should be int32, int64 or compatible. "
                        "Instead got %s." % target.dtype)
     # sparse_softmax_cross_entropy_with_logits requires [batch_size] target.
     if len(target.get_shape()) == 2:
       target = array_ops.squeeze(target, squeeze_dims=[1])
     loss_vec = nn.sparse_softmax_cross_entropy_with_logits(
         logits, target)
   return loss_vec

def sparse_softmax_cross_entropy(labels, logits, weights=1.0, scope=None,
                                 loss_collection=ops.GraphKeys.LOSSES):
  """Cross-entropy loss using `tf.nn.sparse_softmax_cross_entropy_with_logits`.

  `weights` acts as a coefficient for the loss. If a scalar is provided,
  then the loss is simply scaled by the given value. If `weights` is a
  tensor of shape [`batch_size`], then the loss weights apply to each
  corresponding sample.

  WARNING: `weights` also supports dimensions of 1, but the broadcasting does
  not work as advertised, you'll wind up with weighted sum instead of weighted
  mean for any but the last dimension. This will be cleaned up soon, so please
  do not rely on the current behavior for anything but the shapes documented for
  `weights` below.

  Args:
    labels: [batch_size, 1] or [batch_size] target labels of dtype `int32` or
      `int64` in the range `[0, num_classes)`.
    logits: [batch_size, num_classes] logits outputs of the network .
    weights: Coefficients for the loss. This must be of shape `[batch_size]` or
      `[batch_size, 1]`.
    scope: the scope for the operations performed in computing the loss.
    loss_collection: collection to which the loss will be added.

  Returns:
    A scalar `Tensor` representing the mean loss value.

  Raises:
    ValueError: If the shapes of logits, labels, and weight are incompatible, or
      if `weights` is None.
  """
  with ops.name_scope(scope, "sparse_softmax_cross_entropy_loss",
                      [logits, labels, weights]) as scope:
    labels = array_ops.reshape(labels, shape=[array_ops.shape(labels)[0]])

    losses = nn.sparse_softmax_cross_entropy_with_logits(labels=labels,
                                                         logits=logits,
                                                         name="xentropy")
    # Reshape losses to [batch_size, 1] to be consistent with weights.
    losses = array_ops.reshape(losses, shape=[array_ops.shape(losses)[0], 1])
    return compute_weighted_loss(losses, weights, scope, loss_collection)