Python tensorflow.ReaderBase() Examples

The following are 14 code examples of tensorflow.ReaderBase(). You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may also want to check out all available functions/classes of the module tensorflow , or try the search function .
Example #1
Source File: skip_thoughts_model.py    From DOTA_models with Apache License 2.0 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-self.config.uniform_init_scale,
        maxval=self.config.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None
    self.decode_pre_ids = None
    self.decode_post_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None
    self.decode_pre_mask = None
    self.decode_post_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None
    self.decode_pre_emb = None
    self.decode_post_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The cross entropy losses and corresponding weights of the decoders. Used
    # for evaluation.
    self.target_cross_entropy_losses = []
    self.target_cross_entropy_loss_weights = []

    # The total loss to optimize.
    self.total_loss = None
Example #2
Source File: skip_thoughts_model.py    From yolo_v2 with Apache License 2.0 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-self.config.uniform_init_scale,
        maxval=self.config.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None
    self.decode_pre_ids = None
    self.decode_post_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None
    self.decode_pre_mask = None
    self.decode_post_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None
    self.decode_pre_emb = None
    self.decode_post_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The cross entropy losses and corresponding weights of the decoders. Used
    # for evaluation.
    self.target_cross_entropy_losses = []
    self.target_cross_entropy_loss_weights = []

    # The total loss to optimize.
    self.total_loss = None
Example #3
Source File: skip_thoughts_model.py    From parallax with Apache License 2.0 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
        """Basic setup. The actual TensorFlow graph is constructed in build().

        Args:
          config: Object containing configuration parameters.
          mode: "train", "eval" or "encode".
          input_reader: Subclass of tf.ReaderBase for reading the input serialized
            tf.Example protocol buffers. Defaults to TFRecordReader.

        Raises:
          ValueError: If mode is invalid.
        """
        if mode not in ["train", "eval", "encode"]:
            raise ValueError("Unrecognized mode: %s" % mode)

        self.config = config
        self.mode = mode
        self.reader = input_reader if input_reader else tf.TFRecordReader()

        # Initializer used for non-recurrent weights.
        self.uniform_initializer = tf.random_uniform_initializer(
            minval=-self.config.uniform_init_scale,
            maxval=self.config.uniform_init_scale)

        # Input sentences represented as sequences of word ids. "encode" is the
        # source sentence, "decode_pre" is the previous sentence and
        # "decode_post" is the next sentence.
        # Each is an int64 Tensor with  shape [batch_size, padded_length].
        self.encode_ids = None
        self.decode_pre_ids = None
        self.decode_post_ids = None

        # Boolean masks distinguishing real words (1) from padded words (0).
        # Each is an int32 Tensor with shape [batch_size, padded_length].
        self.encode_mask = None
        self.decode_pre_mask = None
        self.decode_post_mask = None

        # Input sentences represented as sequences of word embeddings.
        # Each is a float32 Tensor with shape
        # [batch_size, padded_length, emb_dim].
        self.encode_emb = None
        self.decode_pre_emb = None
        self.decode_post_emb = None

        # The output from the sentence encoder.
        # A float32 Tensor with shape [batch_size, num_gru_units].
        self.thought_vectors = None

        # The cross entropy losses and corresponding weights of the decoders.
        # Used for evaluation.
        self.target_cross_entropy_losses = []
        self.target_cross_entropy_loss_weights = []

        # The total loss to optimize.
        self.total_loss = None
Example #4
Source File: skip_thoughts_model.py    From Gun-Detector with Apache License 2.0 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-self.config.uniform_init_scale,
        maxval=self.config.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None
    self.decode_pre_ids = None
    self.decode_post_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None
    self.decode_pre_mask = None
    self.decode_post_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None
    self.decode_pre_emb = None
    self.decode_post_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The cross entropy losses and corresponding weights of the decoders. Used
    # for evaluation.
    self.target_cross_entropy_losses = []
    self.target_cross_entropy_loss_weights = []

    # The total loss to optimize.
    self.total_loss = None
Example #5
Source File: s2v_model.py    From S2V with Apache License 2.0 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None, input_queue=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()
    self.input_queue = input_queue

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-FLAGS.uniform_init_scale,
        maxval=FLAGS.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The total loss to optimize.
    self.total_loss = None
Example #6
Source File: skip_thoughts_model.py    From hands-detection with MIT License 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-self.config.uniform_init_scale,
        maxval=self.config.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None
    self.decode_pre_ids = None
    self.decode_post_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None
    self.decode_pre_mask = None
    self.decode_post_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None
    self.decode_pre_emb = None
    self.decode_post_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The cross entropy losses and corresponding weights of the decoders. Used
    # for evaluation.
    self.target_cross_entropy_losses = []
    self.target_cross_entropy_loss_weights = []

    # The total loss to optimize.
    self.total_loss = None
Example #7
Source File: skip_thoughts_model.py    From object_detection_kitti with Apache License 2.0 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-self.config.uniform_init_scale,
        maxval=self.config.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None
    self.decode_pre_ids = None
    self.decode_post_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None
    self.decode_pre_mask = None
    self.decode_post_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None
    self.decode_pre_emb = None
    self.decode_post_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The cross entropy losses and corresponding weights of the decoders. Used
    # for evaluation.
    self.target_cross_entropy_losses = []
    self.target_cross_entropy_loss_weights = []

    # The total loss to optimize.
    self.total_loss = None
Example #8
Source File: skip_thoughts_model.py    From object_detection_with_tensorflow with MIT License 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-self.config.uniform_init_scale,
        maxval=self.config.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None
    self.decode_pre_ids = None
    self.decode_post_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None
    self.decode_pre_mask = None
    self.decode_post_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None
    self.decode_pre_emb = None
    self.decode_post_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The cross entropy losses and corresponding weights of the decoders. Used
    # for evaluation.
    self.target_cross_entropy_losses = []
    self.target_cross_entropy_loss_weights = []

    # The total loss to optimize.
    self.total_loss = None
Example #9
Source File: skip_thoughts_model.py    From HumanRecognition with MIT License 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-self.config.uniform_init_scale,
        maxval=self.config.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None
    self.decode_pre_ids = None
    self.decode_post_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None
    self.decode_pre_mask = None
    self.decode_post_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None
    self.decode_pre_emb = None
    self.decode_post_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The cross entropy losses and corresponding weights of the decoders. Used
    # for evaluation.
    self.target_cross_entropy_losses = []
    self.target_cross_entropy_loss_weights = []

    # The total loss to optimize.
    self.total_loss = None
Example #10
Source File: s2v_model.py    From text_embedding with MIT License 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None, input_queue=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()
    self.input_queue = input_queue

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-FLAGS.uniform_init_scale,
        maxval=FLAGS.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The total loss to optimize.
    self.total_loss = None
Example #11
Source File: skip_thoughts_model.py    From text_embedding with MIT License 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-self.config.uniform_init_scale,
        maxval=self.config.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None
    self.decode_pre_ids = None
    self.decode_post_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None
    self.decode_pre_mask = None
    self.decode_post_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None
    self.decode_pre_emb = None
    self.decode_post_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The cross entropy losses and corresponding weights of the decoders. Used
    # for evaluation.
    self.target_cross_entropy_losses = []
    self.target_cross_entropy_loss_weights = []

    # The total loss to optimize.
    self.total_loss = None
Example #12
Source File: skip_thoughts_model.py    From g-tensorflow-models with Apache License 2.0 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-self.config.uniform_init_scale,
        maxval=self.config.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None
    self.decode_pre_ids = None
    self.decode_post_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None
    self.decode_pre_mask = None
    self.decode_post_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None
    self.decode_pre_emb = None
    self.decode_post_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The cross entropy losses and corresponding weights of the decoders. Used
    # for evaluation.
    self.target_cross_entropy_losses = []
    self.target_cross_entropy_loss_weights = []

    # The total loss to optimize.
    self.total_loss = None
Example #13
Source File: skip_thoughts_model.py    From models with Apache License 2.0 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-self.config.uniform_init_scale,
        maxval=self.config.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None
    self.decode_pre_ids = None
    self.decode_post_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None
    self.decode_pre_mask = None
    self.decode_post_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None
    self.decode_pre_emb = None
    self.decode_post_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The cross entropy losses and corresponding weights of the decoders. Used
    # for evaluation.
    self.target_cross_entropy_losses = []
    self.target_cross_entropy_loss_weights = []

    # The total loss to optimize.
    self.total_loss = None
Example #14
Source File: skip_thoughts_model.py    From multilabel-image-classification-tensorflow with MIT License 4 votes vote down vote up 
def __init__(self, config, mode="train", input_reader=None):
    """Basic setup. The actual TensorFlow graph is constructed in build().

    Args:
      config: Object containing configuration parameters.
      mode: "train", "eval" or "encode".
      input_reader: Subclass of tf.ReaderBase for reading the input serialized
        tf.Example protocol buffers. Defaults to TFRecordReader.

    Raises:
      ValueError: If mode is invalid.
    """
    if mode not in ["train", "eval", "encode"]:
      raise ValueError("Unrecognized mode: %s" % mode)

    self.config = config
    self.mode = mode
    self.reader = input_reader if input_reader else tf.TFRecordReader()

    # Initializer used for non-recurrent weights.
    self.uniform_initializer = tf.random_uniform_initializer(
        minval=-self.config.uniform_init_scale,
        maxval=self.config.uniform_init_scale)

    # Input sentences represented as sequences of word ids. "encode" is the
    # source sentence, "decode_pre" is the previous sentence and "decode_post"
    # is the next sentence.
    # Each is an int64 Tensor with  shape [batch_size, padded_length].
    self.encode_ids = None
    self.decode_pre_ids = None
    self.decode_post_ids = None

    # Boolean masks distinguishing real words (1) from padded words (0).
    # Each is an int32 Tensor with shape [batch_size, padded_length].
    self.encode_mask = None
    self.decode_pre_mask = None
    self.decode_post_mask = None

    # Input sentences represented as sequences of word embeddings.
    # Each is a float32 Tensor with shape [batch_size, padded_length, emb_dim].
    self.encode_emb = None
    self.decode_pre_emb = None
    self.decode_post_emb = None

    # The output from the sentence encoder.
    # A float32 Tensor with shape [batch_size, num_gru_units].
    self.thought_vectors = None

    # The cross entropy losses and corresponding weights of the decoders. Used
    # for evaluation.
    self.target_cross_entropy_losses = []
    self.target_cross_entropy_loss_weights = []

    # The total loss to optimize.
    self.total_loss = None