Python sugartensor.sg_floatx方法代碼示例

# 需要導入模塊: import sugartensor [as 別名]
# 或者: from sugartensor import sg_floatx [as 別名]
def sg_summary_activation(tensor, prefix=None, name=None):
    r"""Register `tensor` to summary report as `activation`

    Args:
      tensor: A `Tensor` to log as activation
      prefix: A `string`. A prefix to display in the tensor board web UI.
      name: A `string`. A name to display in the tensor board web UI.

    Returns:
      None
    """
    # defaults
    prefix = '' if prefix is None else prefix + '/'
    # summary name
    name = prefix + _pretty_name(tensor) if name is None else prefix + name
    # summary statistics
    _scalar(name + '/ratio',
            tf.reduce_mean(tf.cast(tf.greater(tensor, 0), tf.sg_floatx)))
    _histogram(name + '/ratio-h', tensor) 

# 需要導入模塊: import sugartensor [as 別名]
# 或者: from sugartensor import sg_floatx [as 別名]
def glorot_uniform(name, shape, scale=1, dtype=tf.sg_floatx, summary=True, regularizer=None, trainable=True):
    r"""See [Glorot & Bengio. 2010.](http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf)

    Args:
      name: The name of new variable
      shape: A tuple/list of integers.
      scale: A Python scalar. Scale to initialize. Default is 1.
      dtype: The data type. Default is float32.
      summary: If True, add this constant to tensor board summary.
      regularizer:  A (Tensor -> Tensor or None) function; the result of applying it on a newly created variable
        will be added to the collection tf.GraphKeys.REGULARIZATION_LOSSES and can be used for regularization
      trainable: If True, add this constant to trainable collection. Default is True.

    Returns:
      A `Variable`.

    """
    fin, fout = _get_fans(shape)
    s = np.sqrt(6. * scale / (fin + fout))
    return uniform(name, shape, s, dtype, summary, regularizer, trainable) 

# 需要導入模塊: import sugartensor [as 別名]
# 或者: from sugartensor import sg_floatx [as 別名]
def sg_emb(**kwargs):
    r"""Returns a look-up table for embedding.
    
    kwargs:
      name: A name for the layer.
      emb: A 2-D array (optional). 
        If None, the resulting tensor should have the shape of 
        `[vocabulary size, embedding dimension size]`.
        Note that its first row is filled with 0's associated with padding.
      in_dim: A positive `integer`. The size of input dimension.
      dim: A positive `integer`. The size of output dimension.
      voca_size: A positive integer. The size of vocabulary.
      summary: If True, summaries are added. The default is True.

    Returns:
      A 2-D `Tensor` of float32.
    """
    opt = tf.sg_opt(kwargs)
    assert opt.name is not None, 'name is mandatory.'

    if opt.emb is None:
        # initialize embedding matrix
        assert opt.voca_size is not None, 'voca_size is mandatory.'
        assert opt.dim is not None, 'dim is mandatory.'
        w = tf.sg_initializer.he_uniform(opt.name, (opt.voca_size - 1, opt.dim), summary=opt.summary)
    else:
        # use given embedding matrix
        w = tf.sg_initializer.external(opt.name, value=opt.emb, summary=opt.summary)

    # 1st row should be zero and not be updated by backprop because of zero padding.
    emb = tf.concat([tf.zeros((1, opt.dim), dtype=tf.sg_floatx), w], 0)

    return emb


# layer normalization for rnn 

# 需要導入模塊: import sugartensor [as 別名]
# 或者: from sugartensor import sg_floatx [as 別名]
def sg_hinge(tensor, opt):
    r"""Returns hinge loss between `tensor` and `target`.
    
    Args:
      tensor: A `Tensor`.
      opt:
        target: A `Tensor`. Labels.
        margin: An int. Maximum margin. Default is 1.
        name: A `string`. A name to display in the tensor board web UI.
      
    Returns:
      A `Tensor`.
    
    For example,
    
    ```
    tensor = [[30, 10, 40], [13, 30, 42]]
    target = [[0, 0, 1], [0, 1, 0]]
    tensor.sg_hinge(target=target, one_hot=True) =>     [[ 1.  1.  0.]
                                                         [ 1.  0.  1.]]
    ```
    """
    assert opt.target is not None, 'target is mandatory.'

    # default margin
    opt += tf.sg_opt(margin=1)

    # reshape target
    shape = tensor.get_shape().as_list()
    broadcast_shape = [-1] + [1] * (len(shape) - 2) + [shape[-1]]
    target = tf.cast(tf.reshape(opt.target, broadcast_shape), tf.sg_floatx)
    
    # hinge loss
    out = tf.identity(tf.maximum(opt.margin - target * tensor, 0), 'hinge')

    # add summary
    tf.sg_summary_loss(out, name=opt.name)

    return out 

# 需要導入模塊: import sugartensor [as 別名]
# 或者: from sugartensor import sg_floatx [as 別名]
def constant(name, shape, value=0, dtype=tf.sg_floatx, summary=True, regularizer=None, trainable=True):
    r"""Creates a tensor variable of which initial values are `value` and shape is `shape`.

    Args:
      name: The name of new variable.
      shape: A tuple/list of integers or an integer. 
        If shape is an integer, it is converted to a list.
      value: A Python scalar. All elements of the initialized variable
        will be set to this value. Default is 0.
      dtype: The data type. Only floating point types are supported. Default is float32.
      summary: If True, add this constant to tensor board summary.
      regularizer:  A (Tensor -> Tensor or None) function; the result of applying it on a newly created variable
        will be added to the collection tf.GraphKeys.REGULARIZATION_LOSSES and can be used for regularization
      trainable: If True, add this constant to trainable collection. Default is True.

    Returns:
      A `Variable`.

    """
    shape = shape if isinstance(shape, (tuple, list)) else [shape]
    x = tf.get_variable(name, shape, dtype=dtype,
                        initializer=tf.constant_initializer(value),
                        regularizer=regularizer, trainable=trainable)
    # add summary
    if summary:
        tf.sg_summary_param(x)
    return x 

# 需要導入模塊: import sugartensor [as 別名]
# 或者: from sugartensor import sg_floatx [as 別名]
def uniform(name, shape, scale=0.05, dtype=tf.sg_floatx, summary=True, regularizer=None, trainable=True):
    r"""Creates a tensor variable of which initial values are 
    random numbers based on uniform distribution.
    
    Note that the default value of `scale` (=0.05) is different from 
    the min/max values (=0.0, 1.0) of tf.random_uniform_initializer.
    
    Args:
      name: The name of the new variable.
      shape: A tuple/list of integers or an integer. 
        If shape is an integer, it's converted to a list.
      scale: A Python scalar. All initial values should be in range `[-scale, scale)`. Default is .05.
      dtype: The data type. Only floating point types are supported. Default is float32.
      summary: If True, add this constant to tensor board summary.
      regularizer:  A (Tensor -> Tensor or None) function; the result of applying it on a newly created variable
        will be added to the collection tf.GraphKeys.REGULARIZATION_LOSSES and can be used for regularization
      trainable: If True, add this constant to trainable collection. Default is True.

    Returns:
      A `Variable`.
    """
    shape = shape if isinstance(shape, (tuple, list)) else [shape]
    x = tf.get_variable(name, shape, dtype=dtype,
                        initializer=tf.random_uniform_initializer(minval=-scale, maxval=scale),
                        regularizer=regularizer, trainable=trainable)
    # add summary
    if summary:
        tf.sg_summary_param(x)
    return x 

# 需要導入模塊: import sugartensor [as 別名]
# 或者: from sugartensor import sg_floatx [as 別名]
def identity(name, dim, scale=1, dtype=tf.sg_floatx, summary=True, regularizer=None, trainable=True):
    r"""Creates a tensor variable of which initial values are of
    an identity matrix.
    
    Note that the default value of `scale` (=0.05) is different from 
    the min/max values (=0.0, 1.0) of tf.random_uniform_initializer.
    
    For example,
    
    ```
    identity("identity", 3, 2) =>
    [[2. 0. 0.]
     [0. 2. 0.]
     [0. 0. 2.]]
    ```
    
    Args:
      name: The name of new variable.
      dim: An int. The size of the first and second dimension of the output tensor.
      scale: A Python scalar. The value on the diagonal.
      dtype: The type of the elements of the resulting tensor.
      summary: If True, add this constant to tensor board summary.
      regularizer:  A (Tensor -> Tensor or None) function; the result of applying it on a newly created variable
        will be added to the collection tf.GraphKeys.REGULARIZATION_LOSSES and can be used for regularization
      trainable: If True, add this constant to trainable collection. Default is True.

    Returns:
      A 2-D `Variable`.
    """
    x = tf.get_variable(name,
                        initializer=tf.constant(np.eye(dim) * scale, dtype=dtype),
                        regularizer=regularizer, trainable=trainable)
    # add summary
    if summary:
        tf.sg_summary_param(x)
    return x 

# 需要導入模塊: import sugartensor [as 別名]
# 或者: from sugartensor import sg_floatx [as 別名]
def orthogonal(name, shape, scale=1.1, dtype=tf.sg_floatx, summary=True, regularizer=None, trainable=True):
    r"""Creates a tensor variable of which initial values are of
    an orthogonal ndarray.
    
    See [Saxe et al. 2014.](http://arxiv.org/pdf/1312.6120.pdf)
    
    Args:
      name: The name of new variable.
      shape: A tuple/list of integers. 
      scale: A Python scalar.
      dtype: Either float32 or float64.
      summary: If True, add this constant to tensor board summary.
      regularizer:  A (Tensor -> Tensor or None) function; the result of applying it on a newly created variable
        will be added to the collection tf.GraphKeys.REGULARIZATION_LOSSES and can be used for regularization
      trainable: If True, add this constant to trainable collection. Default is True.

    Returns:
      A `Variable`.
    """
    flat_shape = (shape[0], np.prod(shape[1:]))
    a = np.random.normal(0.0, 1.0, flat_shape)
    u, _, v = np.linalg.svd(a, full_matrices=False)
    # pick the one with the correct shape
    q = u if u.shape == flat_shape else v
    q = q.reshape(shape)
    # create variable
    x = tf.get_variable(name,
                        initializer=tf.constant(scale * q[:shape[0], :shape[1]], dtype=dtype),
                        regularizer=regularizer, trainable=trainable)
    # add summary
    if summary:
        tf.sg_summary_param(x)
    return x 

# 需要導入模塊: import sugartensor [as 別名]
# 或者: from sugartensor import sg_floatx [as 別名]
def external(name, value, dtype=tf.sg_floatx, summary=True, regularizer=None, trainable=True):
    r"""Creates a tensor variable of which initial values are `value`.
    
    For example,
    
    ```
    external("external", [3,3,1,2])
    => [3. 3. 1. 2.]
    ```
    
    Args:
      name: The name of new variable.
      value: A constant value (or list) of output type `dtype`.
      dtype: The type of the elements of the resulting tensor.
      summary: If True, add this constant to tensor board summary.
      regularizer:  A (Tensor -> Tensor or None) function; the result of applying it on a newly created variable
        will be added to the collection tf.GraphKeys.REGULARIZATION_LOSSES and can be used for regularization
      trainable: If True, add this constant to trainable collection. Default is True.

    Returns:
      A `Variable`. Has the same contents as `value` of `dtype`. 
    """
    # create variable
    x = tf.get_variable(name,
                        initializer=tf.constant(value, dtype=dtype),
                        regularizer=regularizer, trainable=trainable)
    # add summary
    if summary:
        tf.sg_summary_param(x)
    return x 

# 需要導入模塊: import sugartensor [as 別名]
# 或者: from sugartensor import sg_floatx [as 別名]
def sg_float(tensor, opt):
    r"""Casts a tensor to floatx.
    
    See `tf.cast()` in tensorflow.

    Args:
      tensor: A `Tensor` or `SparseTensor` (automatically given by chain).
      opt:
        name : If provided, it replaces current tensor's name

    Returns:
      A `Tensor` or `SparseTensor` with same shape as `tensor`.
    """
    return tf.cast(tensor, tf.sg_floatx, name=opt.name) 

# 需要導入模塊: import sugartensor [as 別名]
# 或者: from sugartensor import sg_floatx [as 別名]
def __init__(self, batch_size=16, set_name='train'):

        # load meta file
        label, mfcc_file = [], []
        with open(_data_path + 'preprocess/meta/%s.csv' % set_name) as csv_file:
            reader = csv.reader(csv_file, delimiter=',')
            for row in reader:
                # mfcc file
                mfcc_file.append(_data_path + 'preprocess/mfcc/' + row[0] + '.npy')
                # label info ( convert to string object for variable-length support )
                label.append(np.asarray(row[1:], dtype=np.int).tostring())

        # to constant tensor
        label_t = tf.convert_to_tensor(label)
        mfcc_file_t = tf.convert_to_tensor(mfcc_file)

        # create queue from constant tensor
        label_q, mfcc_file_q \
            = tf.train.slice_input_producer([label_t, mfcc_file_t], shuffle=True)

        # create label, mfcc queue
        label_q, mfcc_q = _load_mfcc(source=[label_q, mfcc_file_q],
                                     dtypes=[tf.sg_intx, tf.sg_floatx],
                                     capacity=256, num_threads=64)

        # create batch queue with dynamic pad
        batch_queue = tf.train.batch([label_q, mfcc_q], batch_size,
                                     shapes=[(None,), (20, None)],
                                     num_threads=64, capacity=batch_size*32,
                                     dynamic_pad=True)

        # split data
        self.label, self.mfcc = batch_queue
        # batch * time * dim
        self.mfcc = self.mfcc.sg_transpose(perm=[0, 2, 1])
        # calc total batch count
        self.num_batch = len(label) // batch_size

        # print info
        tf.sg_info('%s set loaded.(total data=%d, total batch=%d)'
                   % (set_name.upper(), len(label), self.num_batch))