|
@@ 6-60 (lines=55) @@
|
| 3 |
|
from . import variable_summary |
| 4 |
|
|
| 5 |
|
|
| 6 |
|
class HiddenLayer: |
| 7 |
|
""" Typical hidden layer for Multi-layer perceptron |
| 8 |
|
User is allowed to specify the non-linearity activation function. |
| 9 |
|
|
| 10 |
|
Args: |
| 11 |
|
n_in (:obj:`int`): Number of input cells. |
| 12 |
|
n_out (:obj:`int`): Number of output cells. |
| 13 |
|
name (:obj:`str`): Name of the hidden layer. |
| 14 |
|
x (:class:`tensorflow.placeholder`): Input tensor. |
| 15 |
|
W (:class:`tensorflow.Variable`): Weight matrix. |
| 16 |
|
b (:class:`tensorflow.Variable`): Bias matrix. |
| 17 |
|
activation_fn: Activation function used in this hidden layer. |
| 18 |
|
Common values :method:`tensorflow.sigmoid` for ``sigmoid`` function, :method:`tensorflow.tanh` for ``tanh`` |
| 19 |
|
function, :method:`tensorflow.relu` for RELU. |
| 20 |
|
|
| 21 |
|
Attributes: |
| 22 |
|
n_in (:obj:`int`): Number of inputs into this layer. |
| 23 |
|
n_out (:obj:`int`): Number of outputs out of this layer. |
| 24 |
|
name (:obj:`str`): Name of the hidden layer. |
| 25 |
|
x (:class:`tensorflow.placeholder`): Tensorflow placeholder or tensor that represents the input of this layer. |
| 26 |
|
W (:class:`tensorflow.Variable`): Weight matrix of current layer. |
| 27 |
|
b (:class:`tensorflow.Variable`): Bias matrix of current layer. |
| 28 |
|
variables (:obj:`list` of :class:`tensorflow.Variable`): variables of current layer. |
| 29 |
|
logits (:obj:`tensorflow.Tensor`): Tensorflow tensor of linear logits computed in current layer. |
| 30 |
|
y (:class:`tensorflow.Tensor`): Tensorflow tensor represents the output function of this layer. |
| 31 |
|
summaries (:obj:`list`): List of Tensorflow summary buffer. |
| 32 |
|
""" |
| 33 |
|
def __init__(self, n_in, n_out, name, x=None, W=None, b=None, activation_fn=tf.sigmoid): |
| 34 |
|
self.n_in = n_in |
| 35 |
|
self.n_out = n_out |
| 36 |
|
self.name = name |
| 37 |
|
with tf.name_scope(name): |
| 38 |
|
if x is None: |
| 39 |
|
self.x = tf.placeholder(tf.float32, shape=[None, n_in]) |
| 40 |
|
else: |
| 41 |
|
self.x = x |
| 42 |
|
if W is None: |
| 43 |
|
self.W = tf.Variable( |
| 44 |
|
tf.truncated_normal(shape=[n_in, n_out],stddev=1.0/math.sqrt(float(n_in))), |
| 45 |
|
name='weights' |
| 46 |
|
) |
| 47 |
|
else: |
| 48 |
|
self.W = W |
| 49 |
|
if b is None: |
| 50 |
|
self.b = tf.Variable(tf.zeros(shape=[n_out]), name='biases') |
| 51 |
|
else: |
| 52 |
|
self.b = b |
| 53 |
|
self.variables = [self.W, self.b] |
| 54 |
|
self.logits = tf.matmul(self.x, self.W) + self.b |
| 55 |
|
self.y = activation_fn(self.logits, name='activations') |
| 56 |
|
self.summaries = [] |
| 57 |
|
self.summaries += variable_summary(self.W, tag=name + '/weights') |
| 58 |
|
self.summaries += variable_summary(self.b, tag=name + '/bias') |
| 59 |
|
self.summaries.append(tf.summary.histogram(name + '/pre_act', self.logits)) |
| 60 |
|
self.summaries.append(tf.summary.histogram(name + '/act', self.y)) |
| 61 |
|
|
| 62 |
|
|
| 63 |
|
class SoftmaxLayer: |
|
@@ 63-112 (lines=50) @@
|
| 60 |
|
self.summaries.append(tf.summary.histogram(name + '/act', self.y)) |
| 61 |
|
|
| 62 |
|
|
| 63 |
|
class SoftmaxLayer: |
| 64 |
|
""" Softmax Layer as multi-class binary classification output layer |
| 65 |
|
|
| 66 |
|
Parameters: |
| 67 |
|
n_in (:obj:`int`): Number of input cells. |
| 68 |
|
n_out (:obj:`int`): Number of output cells. |
| 69 |
|
name (:obj:`str`): Name of the layer. |
| 70 |
|
x (:class:`tensorflow.placeholder`): Input tensor. |
| 71 |
|
W (:class:`tensorflow.Variable`): Weight matrix. |
| 72 |
|
b (:class:`tensorflow.Variable`): Bias matrix. |
| 73 |
|
|
| 74 |
|
Attributes: |
| 75 |
|
n_in (:obj:`int`): Number of inputs into this layer. |
| 76 |
|
n_out (:obj:`int`): Number of outputs out of this layer. |
| 77 |
|
name (:obj:`str`): Name of the hidden layer. |
| 78 |
|
x (:class:`tensorflow.placeholder`): Tensorflow placeholder or tensor that represents the input of this layer. |
| 79 |
|
W (:class:`tensorflow.Variable`): Weight matrix of current layer. |
| 80 |
|
b (:class:`tensorflow.Variable`): Bias matrix of current layer. |
| 81 |
|
variables (:obj:`list` of :class:`tensorflow.Variable`): variables of current layer. |
| 82 |
|
logits (:obj:`tensorflow.Tensor`): Tensorflow tensor of linear logits computed in current layer. |
| 83 |
|
y (:class:`tensorflow.Tensor`): Tensorflow tensor represents the output function of this layer. |
| 84 |
|
""" |
| 85 |
|
def __init__(self, n_in, n_out, name, x=None, W=None, b=None): |
| 86 |
|
self.n_in = n_in |
| 87 |
|
self.n_out = n_out |
| 88 |
|
with tf.name_scope(name): |
| 89 |
|
if x is None: |
| 90 |
|
self.x = tf.placeholder(tf.float32, shape=[None, n_in], name='input-x') |
| 91 |
|
else: |
| 92 |
|
self.x = x |
| 93 |
|
if W is None: |
| 94 |
|
self.W = tf.Variable( |
| 95 |
|
tf.truncated_normal(shape=[n_in, n_out], stddev=1.0/math.sqrt(float(n_in))), |
| 96 |
|
name='weights' |
| 97 |
|
) |
| 98 |
|
else: |
| 99 |
|
self.W = W |
| 100 |
|
if b is None: |
| 101 |
|
self.b = tf.Variable(tf.zeros(shape=[n_out]), name='biases') |
| 102 |
|
else: |
| 103 |
|
self.b = b |
| 104 |
|
self.variables = [self.W, self.b] |
| 105 |
|
self.logits = tf.matmul(self.x, self.W) + self.b |
| 106 |
|
self.name = name |
| 107 |
|
self.y = tf.nn.softmax(self.logits, name='softmax') |
| 108 |
|
self.summaries = [] |
| 109 |
|
self.summaries += variable_summary(self.W, tag=name + '/weights') |
| 110 |
|
self.summaries += variable_summary(self.b, tag=name + '/bias') |
| 111 |
|
self.summaries.append(tf.summary.histogram(name + '/pre_act', self.logits)) |
| 112 |
|
self.summaries.append(tf.summary.histogram(name + '/act', self.y)) |
| 113 |
|
|
| 114 |
|
|
| 115 |
|
class AutoencoderLayer(HiddenLayer): |
TinghuiWang /
pyActLearn
