def one_hot(var, n_classes, dtype=None, name=None):
"""Convert to one-hot encoding.
Parameters
----------
n_classes : int
Number of label to encode
dtype : str
The dtype of the resulting Tensor. Default to floatX
name : str
Name of operation
Returns
-------
Tensor
Tensor with shape ``(var.shape[0], n_classes)``
Notes
-----
The Tensor must be either vector or 2D matrix
"""
if not var.n_dim == 1:
raise ValueError('Tensor must be 1D.')
_dtype = dtype or luchador.get_nn_dtype()
_tensor = tf.one_hot(
var.unwrap(), depth=n_classes, dtype=_dtype, name=name)
return Tensor(tensor=_tensor, name=name)
def one_hot(var, n_classes, dtype=None, name=None):
"""Implement ``one_hot`` in Tensorflow backend.
See :func:`luchador.nn.ops.one_hot` for the detail.
"""
_dtype = dtype or luchador.get_nn_dtype()
_tensor = tf.one_hot(
var.unwrap(), depth=n_classes, dtype=_dtype, name=name)
return Tensor(tensor=_tensor, name=name)
def _build(self, input_tensor):
dtype = input_tensor.dtype
tensor = input_tensor.unwrap()
if 'int' in input_tensor.dtype:
dtype = luchador.get_nn_dtype()
tensor = tf.cast(tensor, dtype)
if self._denom is None:
self._instantiate_denominator(dtype)
output = tf.truediv(tensor, self._denom, 'ouptut')
return Tensor(output, name='output')
def _build(self, input_tensor):
dtype = input_tensor.dtype
tensor = input_tensor.unwrap()
if 'int' in input_tensor.dtype:
dtype = luchador.get_nn_dtype()
tensor = tf.cast(tensor, dtype)
if self._denom is None:
self._instantiate_denominator(dtype)
output = tf.truediv(tensor, self._denom, 'ouptut')
return Tensor(output, name='output')
def get_variable(
name, shape=None, dtype=None,
initializer=None, regularizer=None, trainable=True, **kwargs):
"""Create Variable with the given configuration or retrieve existing one
This function works mostly same as tf.get_variable, except when retrieving
existing Variable, you only need name and need not to give shape and dtype.
Mapping from name to VariableWrapper is internally cached so that you can
retrieve variable with only name.
Parameters
----------
name : str
Name of Variable to create or retrieve
shape : list
Used to create new Variable. Ignored when retrieving one
dtype : str
Used to create new Variable. Ignored when retrieving one
initializer : luchador.nn.Initializer or tf.Initializer
Initializer object
kwargs
Other arguments passed to ``tf.get_variable``
See
https://www.tensorflow.org/versions/master/api_docs/python/state_ops.html#get_variable
"""
if isinstance(initializer, base_initializer.BaseInitializer):
initializer = initializer.unwrap()
scope = tf.get_variable_scope()
if scope.reuse:
name = '{}/{}'.format(scope.name, name) if scope.name else name
var = base_wrapper.retrieve_variable(name)
if var is None:
raise ValueError(
'Variable {} does not exist, disallowed. '
'Did you mean to set reuse=None in VarScope?'
.format(name)
)
return var
else:
dtype = dtype or luchador.get_nn_dtype()
variable = tf.get_variable(
name, shape=shape, dtype=dtype, initializer=initializer,
regularizer=regularizer, trainable=trainable, **kwargs)
return Variable(variable, trainable=trainable)
def make_variable(name,
shape,
dtype=None,
initializer=None,
regularizer=None,
trainable=True,
**kwargs):
"""Create Variable with the given configuration
Parameters
----------
name : str
Name of Variable to create.
shape : list
Used to create new Variable.
dtype : str
Used to create new Variable.
initializer : luchador.nn.Initializer
Initializer object
kwargs
Other arguments passed to ``tf.get_variable``
See
https://www.tensorflow.org/versions/master/api_docs/python/state_ops.html#get_variable
"""
dtype = dtype or luchador.get_nn_dtype()
if not initializer:
initializer = fetch_initializer('NormalInitializer')(dtype=dtype)
if isinstance(initializer, BaseInitializer):
initializer = initializer.unwrap()
return Variable(tf.get_variable(name,
shape=shape,
dtype=dtype,
initializer=initializer,
regularizer=regularizer,
trainable=trainable,
**kwargs),
name=name,
trainable=trainable)
def __init__(self, shape, name=None, dtype=None):
"""Creates Input object which wraps placeholder
Parameters
----------
shape : list
The shape of the resulting object.
name : str
The name of the resulting object.
dtype : NumPy dtype or None
If None, default dtype is used
"""
_dtype = dtype or luchador.get_nn_dtype()
tensor = tf.placeholder(dtype=_dtype, shape=shape, name=name)
name = _prefix_with_scope(name)
dtype = _get_dtype_str(tensor)
super(Input, self).__init__(
tensor=tensor, shape=shape, name=name, dtype=dtype)
def __init__(self, shape, name=None, dtype=None):
"""Creates Input object which wraps placeholder
Parameters
----------
shape : list
The shape of the resulting object.
name : str
The name of the resulting object.
dtype : NumPy dtype or None
If None, default dtype is used
"""
_dtype = dtype or luchador.get_nn_dtype()
tensor = tf.placeholder(dtype=_dtype, shape=shape, name=name)
name = _prefix_with_scope(name)
dtype = _get_dtype_str(tensor)
super(Input, self).__init__(
tensor=tensor, shape=shape, name=name, dtype=dtype)
def _write_data_to_file(file_, data):
for key, value in data.items():
_LG.debug(' Saving: %10s %24s %s', value.dtype, value.shape, key)
if key in file_:
del file_[key]
chunks = None if value.size == 1 else True
file_.create_dataset(key, data=value, chunks=chunks)
if 'LUCHADOR_NN_BACKEND' not in file_:
data = np.string_(luchador.get_nn_backend())
file_.create_dataset('LUCHADOR_NN_BACKEND', data=data, dtype='S10')
if 'LUCHADOR_NN_CONV_FORMAT' not in file_:
data = np.string_(luchador.get_nn_conv_format())
file_.create_dataset('LUCHADOR_NN_CONV_FORMAT', data=data, dtype='S4')
if 'LUCHADOR_NN_DTYPE' not in file_:
data = np.string_(luchador.get_nn_dtype())
file_.create_dataset('LUCHADOR_NN_DTYPE', data=data, dtype='S10')
if 'LUCHADOR_VERSION' not in file_:
data = np.string_(luchador.__version__)
file_.create_dataset('LUCHADOR_VERSION', data=data)
file_.flush()
def _write_data_to_file(file_, data):
for key, value in data.items():
_LG.debug(' Saving: %10s %24s %s', value.dtype, value.shape, key)
if key in file_:
del file_[key]
chunks = None if value.size == 1 else True
file_.create_dataset(key, data=value, chunks=chunks)
if 'LUCHADOR_NN_BACKEND' not in file_:
data = np.string_(luchador.get_nn_backend())
file_.create_dataset('LUCHADOR_NN_BACKEND', data=data, dtype='S10')
if 'LUCHADOR_NN_CONV_FORMAT' not in file_:
data = np.string_(luchador.get_nn_conv_format())
file_.create_dataset('LUCHADOR_NN_CONV_FORMAT', data=data, dtype='S4')
if 'LUCHADOR_NN_DTYPE' not in file_:
data = np.string_(luchador.get_nn_dtype())
file_.create_dataset('LUCHADOR_NN_DTYPE', data=data, dtype='S10')
if 'LUCHADOR_VERSION' not in file_:
data = np.string_(luchador.__version__)
file_.create_dataset('LUCHADOR_VERSION', data=data)
file_.flush()
def make_variable(
name, shape, dtype=None,
initializer=None, regularizer=None, trainable=True, **kwargs):
"""Create Variable with the given configuration
Parameters
----------
name : str
Name of Variable to create.
shape : list
Used to create new Variable.
dtype : str
Used to create new Variable.
initializer : luchador.nn.Initializer
Initializer object
kwargs
Other arguments passed to ``tf.get_variable``
See
https://www.tensorflow.org/versions/master/api_docs/python/state_ops.html#get_variable
"""
dtype = dtype or luchador.get_nn_dtype()
if not initializer:
initializer = fetch_initializer('NormalInitializer')(dtype=dtype)
if isinstance(initializer, BaseInitializer):
initializer = initializer.unwrap()
return Variable(
tf.get_variable(
name, shape=shape, dtype=dtype, initializer=initializer,
regularizer=regularizer, trainable=trainable, **kwargs
), name=name, trainable=trainable
)
def _get_dtype(self):
return tf.as_dtype(self.args['dtype'] or luchador.get_nn_dtype())
def _get_dtype(self):
return tf.as_dtype(self.args['dtype'] or luchador.get_nn_dtype())