tensorlayer3/tensorlayer/layers/inputs.py

78 lines
1.9 KiB
Python

#! /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