def _linear_model_fn(features, labels, mode, head, feature_columns, optimizer,
                     partitioner, config):
  """A model_fn for linear models that use a gradient-based optimizer.

  Args:
    features: dict of `Tensor`.
    labels: `Tensor` of shape `[batch_size, logits_dimension]`.
    mode: Defines whether this is training, evaluation or prediction.
      See `ModeKeys`.
    head: A `Head` instance.
    feature_columns: An iterable containing all the feature columns used by
      the model.
    optimizer: string, `Optimizer` object, or callable that defines the
      optimizer to use for training. If `None`, will use a FTRL optimizer.
    partitioner: Partitioner for variables.
    config: `RunConfig` object to configure the runtime settings.

  Returns:
    An `EstimatorSpec` instance.

  Raises:
    ValueError: mode or params are invalid, or features has the wrong type.
  """
  if not isinstance(features, dict):
    raise ValueError('features should be a dictionary of `Tensor`s. '
                     'Given type: {}'.format(type(features)))

  optimizer = optimizers.get_optimizer_instance(
      optimizer or _get_default_optimizer(feature_columns),
      learning_rate=_LEARNING_RATE)
  num_ps_replicas = config.num_ps_replicas if config else 0

  partitioner = partitioner or (
      partitioned_variables.min_max_variable_partitioner(
          max_partitions=num_ps_replicas,
          min_slice_size=64 << 20))

  with variable_scope.variable_scope(
      'linear',
      values=tuple(six.itervalues(features)),
      partitioner=partitioner):

    logit_fn = _linear_logit_fn_builder(
        units=head.logits_dimension, feature_columns=feature_columns)
    logits = logit_fn(features=features)

    def _train_op_fn(loss):
      """Returns the op to optimize the loss."""
      return optimizer.minimize(
          loss,
          global_step=training_util.get_global_step())

    return head.create_estimator_spec(
        features=features,
        mode=mode,
        labels=labels,
        train_op_fn=_train_op_fn,
        logits=logits)

  def test_object(self):
    class _TestOptimizer(optimizer_lib.Optimizer):

      def __init__(self):
        super(_TestOptimizer, self).__init__(
            use_locking=False, name='TestOptimizer')

    opt = optimizers.get_optimizer_instance(_TestOptimizer())
    self.assertIsInstance(opt, _TestOptimizer)

def _linear_model_fn(features, labels, mode, params, config):
  """A model_fn for linear models that use a gradient-based optimizer.

  Args:
    features: Dict of `Tensor`.
    labels: `Tensor` of shape `[batch_size, logits_dimension]`.
    mode: Defines whether this is training, evaluation or prediction.
      See `ModeKeys`.
    params: A dict of hyperparameters.
      The following hyperparameters are expected:
      * head: A `Head` instance.
      * feature_columns: An iterable containing all the feature columns used by
          the model.
      * optimizer: string, `Optimizer` object, or callable that defines the
          optimizer to use for training. If `None`, will use a FTRL optimizer.
    config: `RunConfig` object to configure the runtime settings.

  Returns:
    An `EstimatorSpec` instance.

  Raises:
    ValueError: If mode or params are invalid.
  """
  head = params['head']
  feature_columns = tuple(params['feature_columns'])
  optimizer = optimizers.get_optimizer_instance(
      params.get('optimizer') or _get_default_optimizer(feature_columns),
      learning_rate=_LEARNING_RATE)
  num_ps_replicas = config.num_ps_replicas if config else 0

  partitioner = params.get('partitioner') or (
      partitioned_variables.min_max_variable_partitioner(
          max_partitions=num_ps_replicas,
          min_slice_size=64 << 20))

  with variable_scope.variable_scope(
      'linear',
      values=tuple(six.itervalues(features)),
      partitioner=partitioner):

    logits = feature_column_lib.linear_model(
        features=features,
        feature_columns=feature_columns,
        units=head.logits_dimension)

    def _train_op_fn(loss):
      """Returns the op to optimize the loss."""
      return optimizer.minimize(
          loss,
          global_step=training_util.get_global_step())

    return head.create_estimator_spec(
        features=features,
        mode=mode,
        labels=labels,
        train_op_fn=_train_op_fn,
        logits=logits)

def _dnn_linear_combined_model_fn(features,
                                  labels,
                                  mode,
                                  head,
                                  linear_feature_columns=None,
                                  linear_optimizer='Ftrl',
                                  dnn_feature_columns=None,
                                  dnn_optimizer='Adagrad',
                                  dnn_hidden_units=None,
                                  dnn_activation_fn=nn.relu,
                                  dnn_dropout=None,
                                  input_layer_partitioner=None,
                                  config=None):
  """Deep Neural Net and Linear combined model_fn.

  Args:
    features: dict of `Tensor`.
    labels: `Tensor` of shape [batch_size, 1] or [batch_size] labels of dtype
      `int32` or `int64` in the range `[0, n_classes)`.
    mode: Defines whether this is training, evaluation or prediction.
      See `ModeKeys`.
    head: A `Head` instance.
    linear_feature_columns: An iterable containing all the feature columns used
      by the Linear model.
    linear_optimizer: string, `Optimizer` object, or callable that defines the
      optimizer to use for training the Linear model. Defaults to the Ftrl
      optimizer.
    dnn_feature_columns: An iterable containing all the feature columns used by
      the DNN model.
    dnn_optimizer: string, `Optimizer` object, or callable that defines the
      optimizer to use for training the DNN model. Defaults to the Adagrad
      optimizer.
    dnn_hidden_units: List of hidden units per DNN layer.
    dnn_activation_fn: Activation function applied to each DNN layer. If `None`,
      will use `tf.nn.relu`.
    dnn_dropout: When not `None`, the probability we will drop out a given DNN
      coordinate.
    input_layer_partitioner: Partitioner for input layer.
    config: `RunConfig` object to configure the runtime settings.

  Returns:
    An `EstimatorSpec` instance.

  Raises:
    ValueError: If both `linear_feature_columns` and `dnn_features_columns`
      are empty at the same time, or `input_layer_partitioner` is missing,
      or features has the wrong type.
  """
  if not isinstance(features, dict):
    raise ValueError('features should be a dictionary of `Tensor`s. '
                     'Given type: {}'.format(type(features)))
  if not linear_feature_columns and not dnn_feature_columns:
    raise ValueError(
        'Either linear_feature_columns or dnn_feature_columns must be defined.')

  num_ps_replicas = config.num_ps_replicas if config else 0
  input_layer_partitioner = input_layer_partitioner or (
      partitioned_variables.min_max_variable_partitioner(
          max_partitions=num_ps_replicas,
          min_slice_size=64 << 20))

  # Build DNN Logits.
  dnn_parent_scope = 'dnn'

  if not dnn_feature_columns:
    dnn_logits = None
  else:
    dnn_optimizer = optimizers.get_optimizer_instance(
        dnn_optimizer, learning_rate=_DNN_LEARNING_RATE)
    _check_no_sync_replicas_optimizer(dnn_optimizer)
    if not dnn_hidden_units:
      raise ValueError(
          'dnn_hidden_units must be defined when dnn_feature_columns is '
          'specified.')
    dnn_partitioner = (
        partitioned_variables.min_max_variable_partitioner(
            max_partitions=num_ps_replicas))
    with variable_scope.variable_scope(
        dnn_parent_scope,
        values=tuple(six.itervalues(features)),
        partitioner=dnn_partitioner):

      dnn_logit_fn = dnn._dnn_logit_fn_builder(  # pylint: disable=protected-access
          units=head.logits_dimension,
          hidden_units=dnn_hidden_units,
          feature_columns=dnn_feature_columns,
          activation_fn=dnn_activation_fn,
          dropout=dnn_dropout,
          input_layer_partitioner=input_layer_partitioner)
      dnn_logits = dnn_logit_fn(features=features, mode=mode)

  linear_parent_scope = 'linear'

  if not linear_feature_columns:
    linear_logits = None
  else:
    linear_optimizer = optimizers.get_optimizer_instance(
        linear_optimizer,
        learning_rate=_linear_learning_rate(len(linear_feature_columns)))
    _check_no_sync_replicas_optimizer(linear_optimizer)
#.........这里部分代码省略.........

def _dnn_model_fn(features,
                  labels,
                  mode,
                  head,
                  hidden_units,
                  feature_columns,
                  optimizer='Adagrad',
                  activation_fn=nn.relu,
                  dropout=None,
                  input_layer_partitioner=None,
                  config=None,
                  tpu_estimator_spec=False):
  """Deep Neural Net model_fn.

  Args:
    features: dict of `Tensor`.
    labels: `Tensor` of shape [batch_size, 1] or [batch_size] labels of
      dtype `int32` or `int64` in the range `[0, n_classes)`.
    mode: Defines whether this is training, evaluation or prediction.
      See `ModeKeys`.
    head: A `head_lib._Head` instance.
    hidden_units: Iterable of integer number of hidden units per layer.
    feature_columns: Iterable of `feature_column._FeatureColumn` model inputs.
    optimizer: String, `tf.Optimizer` object, or callable that creates the
      optimizer to use for training. If not specified, will use the Adagrad
      optimizer with a default learning rate of 0.05.
    activation_fn: Activation function applied to each layer.
    dropout: When not `None`, the probability we will drop out a given
      coordinate.
    input_layer_partitioner: Partitioner for input layer. Defaults
      to `min_max_variable_partitioner` with `min_slice_size` 64 << 20.
    config: `RunConfig` object to configure the runtime settings.
    tpu_estimator_spec: Whether to return a `_TPUEstimatorSpec` or
      or `model_fn.EstimatorSpec` instance.

  Returns:
    An `EstimatorSpec` instance.

  Raises:
    ValueError: If features has the wrong type.
  """
  if not isinstance(features, dict):
    raise ValueError('features should be a dictionary of `Tensor`s. '
                     'Given type: {}'.format(type(features)))

  optimizer = optimizers.get_optimizer_instance(
      optimizer, learning_rate=_LEARNING_RATE)
  num_ps_replicas = config.num_ps_replicas if config else 0

  partitioner = partitioned_variables.min_max_variable_partitioner(
      max_partitions=num_ps_replicas)
  with variable_scope.variable_scope(
      'dnn',
      values=tuple(six.itervalues(features)),
      partitioner=partitioner):
    input_layer_partitioner = input_layer_partitioner or (
        partitioned_variables.min_max_variable_partitioner(
            max_partitions=num_ps_replicas,
            min_slice_size=64 << 20))

    logit_fn = _dnn_logit_fn_builder(
        units=head.logits_dimension,
        hidden_units=hidden_units,
        feature_columns=feature_columns,
        activation_fn=activation_fn,
        dropout=dropout,
        input_layer_partitioner=input_layer_partitioner)
    logits = logit_fn(features=features, mode=mode)

    if tpu_estimator_spec:
      return head._create_tpu_estimator_spec(  # pylint: disable=protected-access
          features=features,
          mode=mode,
          labels=labels,
          optimizer=optimizer,
          logits=logits)
    else:
      return head.create_estimator_spec(
          features=features,
          mode=mode,
          labels=labels,
          optimizer=optimizer,
          logits=logits)

 def train_op_fn(loss):
   opt = optimizers.get_optimizer_instance(
       optimizer, learning_rate=_LEARNING_RATE)
   return opt.minimize(loss, global_step=training_util.get_global_step())

def _rnn_model_fn(features,
                  labels,
                  mode,
                  head,
                  rnn_cell_fn,
                  sequence_feature_columns,
                  context_feature_columns,
                  optimizer='Adagrad',
                  input_layer_partitioner=None,
                  config=None):
  """Recurrent Neural Net model_fn.

  Args:
    features: dict of `Tensor` and `SparseTensor` objects returned from
      `input_fn`.
    labels: `Tensor` of shape [batch_size, 1] or [batch_size] with labels.
    mode: Defines whether this is training, evaluation or prediction.
      See `ModeKeys`.
    head: A `head_lib._Head` instance.
    rnn_cell_fn: A function with one argument, a `tf.estimator.ModeKeys`, and
      returns an object of type `tf.nn.rnn_cell.RNNCell`.
    sequence_feature_columns: Iterable containing `FeatureColumn`s that
      represent sequential model inputs.
    context_feature_columns: Iterable containing `FeatureColumn`s that
      represent model inputs not associated with a specific timestep.
    optimizer: String, `tf.Optimizer` object, or callable that creates the
      optimizer to use for training. If not specified, will use the Adagrad
      optimizer with a default learning rate of 0.05 and gradient clip norm of
      5.0.
    input_layer_partitioner: Partitioner for input layer. Defaults
      to `min_max_variable_partitioner` with `min_slice_size` 64 << 20.
    config: `RunConfig` object to configure the runtime settings.

  Returns:
    An `EstimatorSpec` instance.

  Raises:
    ValueError: If mode or optimizer is invalid, or features has the wrong type.
  """
  if not isinstance(features, dict):
    raise ValueError('features should be a dictionary of `Tensor`s. '
                     'Given type: {}'.format(type(features)))

  # If user does not provide an optimizer instance, use the optimizer specified
  # by the string with default learning rate and gradient clipping.
  if not isinstance(optimizer, optimizer_lib.Optimizer):
    optimizer = optimizers.get_optimizer_instance(
        optimizer, learning_rate=_DEFAULT_LEARNING_RATE)
    optimizer = extenders.clip_gradients_by_norm(optimizer, _DEFAULT_CLIP_NORM)

  num_ps_replicas = config.num_ps_replicas if config else 0
  partitioner = partitioned_variables.min_max_variable_partitioner(
      max_partitions=num_ps_replicas)
  with variable_scope.variable_scope(
      'rnn',
      values=tuple(six.itervalues(features)),
      partitioner=partitioner):
    input_layer_partitioner = input_layer_partitioner or (
        partitioned_variables.min_max_variable_partitioner(
            max_partitions=num_ps_replicas,
            min_slice_size=64 << 20))

    logit_fn = _rnn_logit_fn_builder(
        output_units=head.logits_dimension,
        rnn_cell_fn=rnn_cell_fn,
        sequence_feature_columns=sequence_feature_columns,
        context_feature_columns=context_feature_columns,
        input_layer_partitioner=input_layer_partitioner)
    logits = logit_fn(features=features, mode=mode)

    def _train_op_fn(loss):
      """Returns the op to optimize the loss."""
      return optimizer.minimize(
          loss,
          global_step=training_util.get_global_step())

    return head.create_estimator_spec(
        features=features,
        mode=mode,
        labels=labels,
        train_op_fn=_train_op_fn,
        logits=logits)

 def optimizer_fn():
   return optimizers.get_optimizer_instance('Adagrad', learning_rate=0.05)

def _dnn_model_fn(
    features, labels, mode, head, hidden_units, feature_columns,
    optimizer='Adagrad', activation_fn=nn.relu, dropout=None,
    input_layer_partitioner=None, config=None):
  """Deep Neural Net model_fn.

  Args:
    features: Dict of `Tensor` (depends on data passed to `train`).
    labels: `Tensor` of shape [batch_size, 1] or [batch_size] labels of
      dtype `int32` or `int64` in the range `[0, n_classes)`.
    mode: Defines whether this is training, evaluation or prediction.
      See `ModeKeys`.
    head: A `head_lib._Head` instance.
    hidden_units: Iterable of integer number of hidden units per layer.
    feature_columns: Iterable of `feature_column._FeatureColumn` model inputs.
    optimizer: String, `tf.Optimizer` object, or callable that creates the
      optimizer to use for training. If not specified, will use the Adagrad
      optimizer with a default learning rate of 0.05.
    activation_fn: Activation function applied to each layer.
    dropout: When not `None`, the probability we will drop out a given
      coordinate.
    input_layer_partitioner: Partitioner for input layer. Defaults
      to `min_max_variable_partitioner` with `min_slice_size` 64 << 20.
    config: `RunConfig` object to configure the runtime settings.

  Returns:
    predictions: A dict of `Tensor` objects.
    loss: A scalar containing the loss of the step.
    train_op: The op for training.
  """
  optimizer = optimizers.get_optimizer_instance(
      optimizer, learning_rate=_LEARNING_RATE)
  num_ps_replicas = config.num_ps_replicas if config else 0

  partitioner = partitioned_variables.min_max_variable_partitioner(
      max_partitions=num_ps_replicas)
  with variable_scope.variable_scope(
      'dnn',
      values=tuple(six.itervalues(features)),
      partitioner=partitioner):
    input_layer_partitioner = input_layer_partitioner or (
        partitioned_variables.min_max_variable_partitioner(
            max_partitions=num_ps_replicas,
            min_slice_size=64 << 20))
    with variable_scope.variable_scope(
        'input_from_feature_columns',
        values=tuple(six.itervalues(features)),
        partitioner=input_layer_partitioner):
      net = feature_column_lib.input_layer(
          features=features,
          feature_columns=feature_columns)

    for layer_id, num_hidden_units in enumerate(hidden_units):
      with variable_scope.variable_scope(
          'hiddenlayer_%d' % layer_id,
          values=(net,)) as hidden_layer_scope:
        net = core_layers.dense(
            net,
            units=num_hidden_units,
            activation=activation_fn,
            kernel_initializer=init_ops.glorot_uniform_initializer(),
            name=hidden_layer_scope)
        if dropout is not None and mode == model_fn.ModeKeys.TRAIN:
          net = core_layers.dropout(net, rate=dropout, training=True)
      _add_hidden_layer_summary(net, hidden_layer_scope.name)

    with variable_scope.variable_scope(
        'logits',
        values=(net,)) as logits_scope:
      logits = core_layers.dense(
          net,
          units=head.logits_dimension,
          activation=None,
          kernel_initializer=init_ops.glorot_uniform_initializer(),
          name=logits_scope)
    _add_hidden_layer_summary(logits, logits_scope.name)

    def _train_op_fn(loss):
      """Returns the op to optimize the loss."""
      return optimizer.minimize(
          loss,
          global_step=training_util.get_global_step())

    return head.create_estimator_spec(
        features=features,
        mode=mode,
        labels=labels,
        train_op_fn=_train_op_fn,
        logits=logits)

 def test_supported_name_but_learning_rate_none(self):
   with self.assertRaisesRegexp(
       ValueError, 'learning_rate must be specified when opt is string'):
     optimizers.get_optimizer_instance('Adagrad', learning_rate=None)

 def test_object_invalid(self):
   with self.assertRaisesRegexp(
       ValueError, 'The given object is not an Optimizer instance'):
     optimizers.get_optimizer_instance((1, 2, 3))

 def test_sgd(self):
   opt = optimizers.get_optimizer_instance('SGD', learning_rate=0.1)
   self.assertIsInstance(opt, gradient_descent.GradientDescentOptimizer)
   self.assertAlmostEqual(0.1, opt._learning_rate)

 def test_rmsprop(self):
   opt = optimizers.get_optimizer_instance('RMSProp', learning_rate=0.1)
   self.assertIsInstance(opt, rmsprop.RMSPropOptimizer)
   self.assertAlmostEqual(0.1, opt._learning_rate)

 def test_ftrl(self):
   opt = optimizers.get_optimizer_instance('Ftrl', learning_rate=0.1)
   self.assertIsInstance(opt, ftrl.FtrlOptimizer)
   self.assertAlmostEqual(0.1, opt._learning_rate)

 def test_adam(self):
   opt = optimizers.get_optimizer_instance('Adam', learning_rate=0.1)
   self.assertIsInstance(opt, adam.AdamOptimizer)
   self.assertAlmostEqual(0.1, opt._lr)

  def __init__(self,
               periodicities,
               input_window_size,
               output_window_size,
               model_dir=None,
               num_features=1,
               extra_feature_columns=None,
               num_timesteps=10,
               loss=ar_model.ARModel.NORMAL_LIKELIHOOD_LOSS,
               num_units=128,
               optimizer="Adam",
               config=None):
    """Initialize the Estimator.

    Args:
      periodicities: periodicities of the input data, in the same units as the
        time feature (for example 24 if feeding hourly data with a daily
        periodicity, or 60 * 24 if feeding minute-level data with daily
        periodicity). Note this can be a single value or a list of values for
        multiple periodicities.
      input_window_size: Number of past time steps of data to look at when doing
        the regression.
      output_window_size: Number of future time steps to predict. Note that
        setting this value to > 1 empirically seems to give a better fit.
      model_dir: Directory to save model parameters, graph and etc. This can
        also be used to load checkpoints from the directory into a estimator to
        continue training a previously saved model.
      num_features: The dimensionality of the time series (default value is one
        for univariate, more than one for multivariate).
      extra_feature_columns: A list of `tf.feature_column`s (for example
        `tf.feature_column.embedding_column`) corresponding to features which
        provide extra information to the model but are not part of the series to
        be predicted.
      num_timesteps: Number of buckets into which to divide (time %
        periodicity). This value multiplied by the number of periodicities is
        the number of time features added to the model.
      loss: Loss function to use for training. Currently supported values are
        SQUARED_LOSS and NORMAL_LIKELIHOOD_LOSS. Note that for
        NORMAL_LIKELIHOOD_LOSS, we train the covariance term as well. For
        SQUARED_LOSS, the evaluation loss is reported based on un-scaled
        observations and predictions, while the training loss is computed on
        normalized data.
      num_units: The size of the hidden state in the encoder and decoder LSTM
        cells.
      optimizer: string, `tf.compat.v1.train.Optimizer` object, or callable that
        defines the optimizer algorithm to use for training. Defaults to the
        Adam optimizer with a learning rate of 0.01.
      config: Optional `estimator.RunConfig` object to configure the runtime
        settings.
    """
    optimizer = optimizers.get_optimizer_instance(optimizer, learning_rate=0.01)
    model = ar_model.ARModel(
        periodicities=periodicities,
        input_window_size=input_window_size,
        output_window_size=output_window_size,
        num_features=num_features,
        exogenous_feature_columns=extra_feature_columns,
        num_time_buckets=num_timesteps,
        loss=loss,
        prediction_model_factory=functools.partial(
            ar_model.LSTMPredictionModel, num_units=num_units))
    state_manager = state_management.FilteringOnlyStateManager()
    super(LSTMAutoRegressor, self).__init__(
        model=model,
        state_manager=state_manager,
        optimizer=optimizer,
        model_dir=model_dir,
        config=config,
        head_type=ts_head_lib.OneShotPredictionHead)

 def test_unsupported_name(self):
   with self.assertRaisesRegexp(
       ValueError, 'Unsupported optimizer name: unsupported_name'):
     optimizers.get_optimizer_instance('unsupported_name', learning_rate=0.1)