#! /usr/bin/python
# -*- coding: utf-8 -*-
import tensorflow as tf
from tensorlayer.layers.core import Layer
from tensorlayer.layers.utils import flatten_reshape
from tensorlayer import tl_logging as logging
from tensorlayer.decorators import deprecated_alias
__all__ = [
'FlattenLayer',
'ReshapeLayer',
'TransposeLayer',
]
[docs]class FlattenLayer(Layer):
"""A layer that reshapes high-dimension input into a vector.
Then we often apply DenseLayer, RNNLayer, ConcatLayer and etc on the top of a flatten layer.
[batch_size, mask_row, mask_col, n_mask] ---> [batch_size, mask_row * mask_col * n_mask]
Parameters
----------
prev_layer : :class:`Layer`
Previous layer.
name : str
A unique layer name.
Examples
--------
>>> import tensorflow as tf
>>> import tensorlayer as tl
>>> x = tf.placeholder(tf.float32, shape=[None, 28, 28, 1])
>>> net = tl.layers.InputLayer(x, name='input')
>>> net = tl.layers.FlattenLayer(net, name='flatten')
[?, 784]
"""
@deprecated_alias(layer='prev_layer', end_support_version=1.9) # TODO remove this line for the 1.9 release
def __init__(self, prev_layer, name='flatten'):
super(FlattenLayer, self).__init__(prev_layer=prev_layer, name=name)
_out = flatten_reshape(self.inputs, name=name)
self.n_units = int(_out.get_shape()[-1])
logging.info("FlattenLayer %s: %d" % (self.name, self.n_units))
self.outputs = _out
self._add_layers(self.outputs)
[docs]class ReshapeLayer(Layer):
"""A layer that reshapes a given tensor.
Parameters
----------
prev_layer : :class:`Layer`
Previous layer
shape : tuple of int
The output shape, see ``tf.reshape``.
name : str
A unique layer name.
Examples
--------
>>> import tensorflow as tf
>>> import tensorlayer as tl
>>> x = tf.placeholder(tf.float32, shape=(None, 784))
>>> net = tl.layers.InputLayer(x, name='input')
>>> net = tl.layers.ReshapeLayer(net, [-1, 28, 28, 1], name='reshape')
>>> print(net.outputs)
(?, 28, 28, 1)
"""
@deprecated_alias(layer='prev_layer', end_support_version=1.9) # TODO remove this line for the 1.9 release
def __init__(self, prev_layer, shape, name='reshape'):
super(ReshapeLayer, self).__init__(prev_layer=prev_layer, name=name)
if not shape:
raise ValueError("Shape list can not be empty")
self.outputs = tf.reshape(self.inputs, shape=shape, name=name)
self._add_layers(self.outputs)
logging.info("ReshapeLayer %s: %s" % (self.name, self.outputs.get_shape()))
[docs]class TransposeLayer(Layer):
"""A layer that transposes the dimension of a tensor.
See `tf.transpose() <https://www.tensorflow.org/api_docs/python/tf/transpose>`__ .
Parameters
----------
prev_layer : :class:`Layer`
Previous layer
perm: list of int
The permutation of the dimensions, similar with ``numpy.transpose``.
name : str
A unique layer name.
Examples
----------
>>> import tensorflow as tf
>>> import tensorlayer as tl
>>> x = tf.placeholder(tf.float32, shape=[None, 28, 28, 1])
>>> net = tl.layers.InputLayer(x, name='input')
>>> net = tl.layers.TransposeLayer(net, perm=[0, 1, 3, 2], name='trans')
[None, 28, 1, 28]
"""
@deprecated_alias(layer='prev_layer', end_support_version=1.9) # TODO remove this line for the 1.9 release
def __init__(self, prev_layer, perm, name='transpose'):
if perm is None:
raise AssertionError("The `perm` argument cannot be None")
super(TransposeLayer, self).__init__(prev_layer=prev_layer, name=name)
logging.info("TransposeLayer %s: perm: %s" % (self.name, perm))
self.outputs = tf.transpose(self.inputs, perm=perm, name=name)
self._add_layers(self.outputs)