Camadas Personalizadas
======================
Um fator por trás do sucesso do *Deep Learning* é a disponibilidade de
uma ampla gama de camadas que pode ser composto de maneiras criativas
projetar arquiteturas adequadas para uma ampla variedade de tarefas. Por
exemplo, os pesquisadores inventaram camadas especificamente para lidar
com imagens, texto, loop sobre dados sequenciais, e realizando
programação dinâmica. Mais cedo ou mais tarde, você encontrará ou
inventará uma camada que ainda não existe na estrutura de *Deep
Learning*. Nesses casos, você deve construir uma camada personalizada.
Nesta seção, mostramos como.
Camadas Sem Parâmetros
----------------------
Para começar, construímos uma camada personalizada que não possui
parâmetros próprios. Isso deve parecer familiar, se você se lembra de
nosso introdução ao bloco em :numref:`sec_model_construction`. A
seguinte classe ``CenteredLayer`` simplesmente subtrai a média de sua
entrada. Para construí-lo, simplesmente precisamos herdar da classe da
camada base e implementar a função de propagação direta.
.. raw:: html
.. raw:: html
.. code:: python
from mxnet import np, npx
from mxnet.gluon import nn
npx.set_np()
class CenteredLayer(nn.Block):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def forward(self, X):
return X - X.mean()
.. raw:: html
.. raw:: html
.. code:: python
import torch
from torch import nn
from torch.nn import functional as F
class CenteredLayer(nn.Module):
def __init__(self):
super().__init__()
def forward(self, X):
return X - X.mean()
.. raw:: html
.. raw:: html
.. code:: python
import tensorflow as tf
class CenteredLayer(tf.keras.Model):
def __init__(self):
super().__init__()
def call(self, inputs):
return inputs - tf.reduce_mean(inputs)
.. raw:: html
.. raw:: html
.. raw:: html
.. code:: python
layer = CenteredLayer()
layer(np.array([1, 2, 3, 4, 5]))
.. parsed-literal::
:class: output
array([-2., -1., 0., 1., 2.])
.. raw:: html
.. raw:: html
.. code:: python
layer = CenteredLayer()
layer(torch.FloatTensor([1, 2, 3, 4, 5]))
.. parsed-literal::
:class: output
tensor([-2., -1., 0., 1., 2.])
.. raw:: html
.. raw:: html
.. code:: python
layer = CenteredLayer()
layer(tf.constant([1, 2, 3, 4, 5]))
.. parsed-literal::
:class: output
.. raw:: html
.. raw:: html
.. raw:: html
.. code:: python
net = nn.Sequential()
net.add(nn.Dense(128), CenteredLayer())
net.initialize()
.. raw:: html
.. raw:: html
.. code:: python
net = nn.Sequential(nn.Linear(8, 128), CenteredLayer())
.. raw:: html
.. raw:: html
.. code:: python
net = tf.keras.Sequential([tf.keras.layers.Dense(128), CenteredLayer()])
.. raw:: html
.. raw:: html
.. raw:: html
.. code:: python
Y = net(np.random.uniform(size=(4, 8)))
Y.mean()
.. parsed-literal::
:class: output
array(3.783498e-10)
.. raw:: html
.. raw:: html
.. code:: python
Y = net(torch.rand(4, 8))
Y.mean()
.. parsed-literal::
:class: output
tensor(-4.1910e-09, grad_fn=)
.. raw:: html
.. raw:: html
.. code:: python
Y = net(tf.random.uniform((4, 8)))
tf.reduce_mean(Y)
.. parsed-literal::
:class: output
.. raw:: html
.. raw:: html
.. raw:: html
.. code:: python
class MyDense(nn.Block):
def __init__(self, units, in_units, **kwargs):
super().__init__(**kwargs)
self.weight = self.params.get('weight', shape=(in_units, units))
self.bias = self.params.get('bias', shape=(units,))
def forward(self, x):
linear = np.dot(x, self.weight.data(ctx=x.ctx)) + self.bias.data(
ctx=x.ctx)
return npx.relu(linear)
.. raw:: html
.. raw:: html
.. code:: python
class MyLinear(nn.Module):
def __init__(self, in_units, units):
super().__init__()
self.weight = nn.Parameter(torch.randn(in_units, units))
self.bias = nn.Parameter(torch.randn(units,))
def forward(self, X):
linear = torch.matmul(X, self.weight.data) + self.bias.data
return F.relu(linear)
.. raw:: html
.. raw:: html
.. code:: python
class MyDense(tf.keras.Model):
def __init__(self, units):
super().__init__()
self.units = units
def build(self, X_shape):
self.weight = self.add_weight(name='weight',
shape=[X_shape[-1], self.units],
initializer=tf.random_normal_initializer())
self.bias = self.add_weight(
name='bias', shape=[self.units],
initializer=tf.zeros_initializer())
def call(self, X):
linear = tf.matmul(X, self.weight) + self.bias
return tf.nn.relu(linear)
.. raw:: html
.. raw:: html
.. raw:: html
.. code:: python
dense = MyDense(units=3, in_units=5)
dense.params
.. parsed-literal::
:class: output
mydense0_ (
Parameter mydense0_weight (shape=(5, 3), dtype=)
Parameter mydense0_bias (shape=(3,), dtype=)
)
.. raw:: html
.. raw:: html
.. code:: python
dense = MyLinear(5, 3)
dense.weight
.. parsed-literal::
:class: output
Parameter containing:
tensor([[-0.4359, -0.8619, -0.4170],
[-0.3191, 0.3732, 0.7508],
[-0.1494, -0.8253, -0.7071],
[ 1.0940, -0.4332, -2.2712],
[ 1.9598, -0.2122, 1.8529]], requires_grad=True)
.. raw:: html
.. raw:: html
.. code:: python
dense = MyDense(3)
dense(tf.random.uniform((2, 5)))
dense.get_weights()
.. parsed-literal::
:class: output
[array([[-0.03506669, 0.01943431, -0.00358615],
[ 0.02064196, -0.07825888, -0.00893933],
[ 0.00577814, 0.0454763 , 0.02938056],
[-0.03217198, -0.05240436, -0.00833275],
[ 0.05816463, 0.02635984, -0.05368067]], dtype=float32),
array([0., 0., 0.], dtype=float32)]
.. raw:: html
.. raw:: html
.. raw:: html
.. code:: python
dense.initialize()
dense(np.random.uniform(size=(2, 5)))
.. parsed-literal::
:class: output
array([[0. , 0.01633355, 0. ],
[0. , 0.01581812, 0. ]])
.. raw:: html
.. raw:: html
.. code:: python
dense(torch.rand(2, 5))
.. parsed-literal::
:class: output
tensor([[1.1498, 0.0000, 0.0000],
[1.9590, 0.0000, 0.0000]])
.. raw:: html
.. raw:: html
.. code:: python
dense(tf.random.uniform((2, 5)))
.. parsed-literal::
:class: output
.. raw:: html
.. raw:: html
.. raw:: html
.. code:: python
net = nn.Sequential()
net.add(MyDense(8, in_units=64),
MyDense(1, in_units=8))
net.initialize()
net(np.random.uniform(size=(2, 64)))
.. parsed-literal::
:class: output
array([[0.06508517],
[0.0615553 ]])
.. raw:: html
.. raw:: html
.. code:: python
net = nn.Sequential(MyLinear(64, 8), MyLinear(8, 1))
net(torch.rand(2, 64))
.. parsed-literal::
:class: output
tensor([[0.],
[0.]])
.. raw:: html
.. raw:: html
.. code:: python
net = tf.keras.models.Sequential([MyDense(8), MyDense(1)])
net(tf.random.uniform((2, 64)))
.. parsed-literal::
:class: output
.. raw:: html
.. raw:: html
.. raw:: html
`Discussão `__
.. raw:: html
.. raw:: html
`Discussão `__
.. raw:: html
.. raw:: html
`Discussão `__
.. raw:: html
.. raw:: html