|
- #! /usr/bin/python
- # -*- coding: utf-8 -*-
-
- import tensorlayer as tl
- from tensorlayer import logging
- from tensorlayer.layers.core import Module
-
- __all__ = ['Input', '_InputLayer']
-
-
- class _InputLayer(Module):
- """
- The :class:`Input` class is the starting layer of a neural network.
-
- Parameters
- ----------
- shape : tuple (int)
- Including batch size.
- dtype: dtype
- The type of input values. By default, tf.float32.
- name : None or str
- A unique layer name.
-
- """
-
- def __init__(self, shape, dtype=tl.float32, name=None, init=None):
- super(_InputLayer, self).__init__(name)
-
- logging.info("Input %s: %s" % (self.name, str(shape)))
- self.shape = shape
- self.dtype = dtype
- self.shape_without_none = [_ if _ is not None else 1 for _ in shape]
- if init is None:
- self.outputs = tl.initializers.ones()(self.shape_without_none, dtype=self.dtype)
- else:
- self.outputs = init(self.shape_without_none, dtype=self.dtype)
- self._built = True
-
- def __repr__(self):
- s = 'Input(shape=%s' % str(self.shape)
- if self.name is not None:
- s += (', name=\'%s\'' % self.name)
- s += ')'
- return s
-
- def __call__(self, *args, **kwargs):
- return self.outputs
-
- def build(self, inputs_shape):
- pass
-
- def forward(self):
- return self.outputs
-
-
- def Input(shape, init=tl.initializers.ones(), dtype=tl.float32, name=None):
- """
- The :class:`Input` class is the starting layer of a neural network.
-
- Parameters
- ----------
- shape : tuple (int)
- Including batch size.
- name : None or str
- A unique layer name.
-
- Examples
- ---------
- With TensorLayer
-
- >>> ni = tl.layers.Input([10, 50, 50, 32], name='input')
- >>> output shape : [10, 50, 50, 32]
-
- """
- input_layer = _InputLayer(shape, dtype=dtype, name=name, init=init)
- outputs = input_layer()
- return outputs
|