日本免费高清视频-国产福利视频导航-黄色在线播放国产-天天操天天操天天操天天操|www.shdianci.com

學(xué)無先后,達(dá)者為師

網(wǎng)站首頁 編程語言 正文

pytorch中model.named_parameters()與model.parameters()解讀_python

作者:不想禿頂還想當(dāng)程序猿 ? 更新時(shí)間: 2022-12-25 編程語言

解讀model.named_parameters()與model.parameters()

model.named_parameters()

迭代打印model.named_parameters()將會(huì)打印每一次迭代元素的名字和param。

model = DarkNet([1, 2, 8, 8, 4])
for name, param in model.named_parameters():
? ? print(name,param.requires_grad)
? ? param.requires_grad = False

輸出結(jié)果為

conv1.weight True
bn1.weight True
bn1.bias True
layer1.ds_conv.weight True
layer1.ds_bn.weight True
layer1.ds_bn.bias True
layer1.residual_0.conv1.weight True
layer1.residual_0.bn1.weight True
layer1.residual_0.bn1.bias True
layer1.residual_0.conv2.weight True
layer1.residual_0.bn2.weight True
layer1.residual_0.bn2.bias True
layer2.ds_conv.weight True
layer2.ds_bn.weight True
layer2.ds_bn.bias True
layer2.residual_0.conv1.weight True
layer2.residual_0.bn1.weight True
layer2.residual_0.bn1.bias True
....

并且可以更改參數(shù)的可訓(xùn)練屬性,第一次打印是True,這是第二次,就是False了

model.parameters()

迭代打印model.parameters()將會(huì)打印每一次迭代元素的param而不會(huì)打印名字,這是它和named_parameters的區(qū)別,兩者都可以用來改變r(jià)equires_grad的屬性。

for index, param in enumerate(model.parameters()):
? ? print(param.shape)

輸出結(jié)果為

torch.Size([32, 3, 3, 3])
torch.Size([32])
torch.Size([32])
torch.Size([64, 32, 3, 3])
torch.Size([64])
torch.Size([64])
torch.Size([32, 64, 1, 1])
torch.Size([32])
torch.Size([32])
torch.Size([64, 32, 3, 3])
torch.Size([64])
torch.Size([64])
torch.Size([128, 64, 3, 3])
torch.Size([128])
torch.Size([128])
torch.Size([64, 128, 1, 1])
torch.Size([64])
torch.Size([64])
torch.Size([128, 64, 3, 3])
torch.Size([128])
torch.Size([128])
torch.Size([64, 128, 1, 1])
torch.Size([64])
torch.Size([64])
torch.Size([128, 64, 3, 3])
torch.Size([128])
torch.Size([128])
torch.Size([256, 128, 3, 3])
torch.Size([256])
torch.Size([256])
torch.Size([128, 256, 1, 1])
....

將兩者結(jié)合進(jìn)行迭代,同時(shí)具有索引,網(wǎng)絡(luò)層名字及param

?? ?for index, (name, param) in zip(enumerate(model.parameters()), model.named_parameters()):
?? ??? ?print(index[0])
?? ??? ?print(name, param.shape)

輸出結(jié)果為

0
conv1.weight torch.Size([32, 3, 3, 3])
1
bn1.weight torch.Size([32])
2
bn1.bias torch.Size([32])
3
layer1.ds_conv.weight torch.Size([64, 32, 3, 3])
4
layer1.ds_bn.weight torch.Size([64])
5
layer1.ds_bn.bias torch.Size([64])
6
layer1.residual_0.conv1.weight torch.Size([32, 64, 1, 1])
7
layer1.residual_0.bn1.weight torch.Size([32])
8
layer1.residual_0.bn1.bias torch.Size([32])
9
layer1.residual_0.conv2.weight torch.Size([64, 32, 3, 3])

state_dict()、named_parameters()和parameters()的區(qū)別

Pytorch中有3個(gè)功能極其類似的方法,分別是model.parameters()、model.named_parameters()和model.state_dict(),下面就來探究一下這三種方法的區(qū)別。

它們的差異主要體現(xiàn)在3方面:

  • 返回值類型不同
  • 存儲(chǔ)的模型參數(shù)的種類不同
  • 返回的值的require_grad屬性不同

測試代碼準(zhǔn)備工作

import torch
import torch.nn as nn
import torch.optim as optim
import random
import os
import numpy as np

def seed_torch(seed=1029):
	random.seed(seed)
	os.environ['PYTHONHASHSEED'] = str(seed) # 為了禁止hash隨機(jī)化,使得實(shí)驗(yàn)可復(fù)現(xiàn)
	np.random.seed(seed)
	torch.manual_seed(seed)
	torch.cuda.manual_seed(seed)
	torch.cuda.manual_seed_all(seed) # if you are using multi-GPU.
	torch.backends.cudnn.benchmark = False
	torch.backends.cudnn.deterministic = True

seed_torch() # 固定隨機(jī)數(shù)

# 定義一個(gè)網(wǎng)絡(luò)
class net(nn.Module):
    def __init__(self, num_class=10):
        super(net, self).__init__()
        self.pool1 = nn.AvgPool1d(2)
        self.bn1 = nn.BatchNorm1d(3)
        self.fc1 = nn.Linear(12, 4)
        

    
    def forward(self, x):
        x = self.pool1(x)
        x = self.bn1(x)
        x = x.reshape(x.size(0), -1)
        x = self.fc1(x)

        return x


# 定義網(wǎng)絡(luò)
model = net()

# 定義loss
loss_fn = nn.CrossEntropyLoss()

# 定義優(yōu)化器
optimizer = optim.SGD(model.parameters(), lr=1e-2)

# 定義訓(xùn)練數(shù)據(jù)
x = torch.randn((3, 3, 8))

兩個(gè)概念

可學(xué)習(xí)參數(shù)

可學(xué)習(xí)參數(shù)也可叫做模型參數(shù),其就是要參與學(xué)習(xí)和更新的,特別注意這里的參數(shù)更新是指在優(yōu)化器的optim.step步驟里更新參數(shù),即需要反向傳播更新的參數(shù)

使用nn.parameter.Parameter()創(chuàng)建的變量是可學(xué)習(xí)參數(shù)(模型參數(shù))

模型中的可學(xué)習(xí)參數(shù)的數(shù)據(jù)類型都是nn.parameter.Parameter

optim.step只能更新nn.parameter.Parameter類型的參數(shù)

nn.parameter.Parameter類型的參數(shù)的特點(diǎn)是默認(rèn)requires_grad=True,也就是說訓(xùn)練過程中需要反向傳播的,就需要使用這個(gè)

示例:

在上述定義的網(wǎng)絡(luò)中,self.fc1層中的參數(shù)(weight和bias)是可學(xué)習(xí)參數(shù),要在訓(xùn)練過程中進(jìn)行學(xué)習(xí)與更新

print(type(model.fc1.weight))
(bbn) jyzhang@admin2-X10DAi:~/test$ python net.py
<class 'torch.nn.parameter.Parameter'>

不可學(xué)習(xí)參數(shù)

不可學(xué)習(xí)參數(shù)不參與學(xué)習(xí)和在優(yōu)化器中的更新,即不需要參與反向傳播

不可學(xué)習(xí)參數(shù)將會(huì)通過Module.register_parameter()注冊在self._buffers中,self._buffers是一個(gè)OrderedDict

舉例:上述定義的模型中,self.bn1層中的參數(shù)running_mean、running_var和num_batches_tracked均是不可學(xué)習(xí)參數(shù)

self.register_parameter('running_mean', None)

存儲(chǔ)在self._buffers中的不可學(xué)習(xí)參數(shù)不能通過optim.step()更新參數(shù),但例如上述的self.bn1層中的不可學(xué)習(xí)參數(shù)也會(huì)更新,其更新是發(fā)生在forward的過程中

示例:

在上述定義的網(wǎng)絡(luò)中,self.bn1層中的參數(shù)(running_mean)是不可學(xué)習(xí)參數(shù)

print(type(model.bn1.running_mean))
(bbn) jyzhang@admin2-X10DAi:~/test$ python net.py
<class 'torch.Tensor'>

named_parameters()

總述

model.named_parameters()返回的是一個(gè)生成器(generator),該生成器中只保存了可學(xué)習(xí)、可被優(yōu)化器更新的參數(shù)的參數(shù)名和具體的參數(shù),可通過循環(huán)迭代打印參數(shù)名和參數(shù)(參見代碼示例一)

該方法可以用來改變可學(xué)習(xí)、可被優(yōu)化器更新參數(shù)的requires_grad屬性,因此可用于鎖住某些層的參數(shù),讓其在訓(xùn)練的時(shí)候不更新參數(shù)(參見代碼示例二)

代碼示例一

# model.named_parameters()的用法
print(type(model.named_parameters()))

for name, param in model.named_parameters():
? ? print(name)
? ? print(param)

結(jié)果

(bbn) jyzhang@admin2-X10DAi:~/test$ python net.py
<class 'generator'>
bn1.weight
Parameter containing:
tensor([1., 1., 1.], requires_grad=True)
bn1.bias
Parameter containing:
tensor([0., 0., 0.], requires_grad=True)
fc1.weight
Parameter containing:
tensor([[ 0.0036, ?0.1960, ?0.2315, -0.2408, ?0.1217, ?0.2579, -0.0676, -0.1880,
? ? ? ? ?-0.2855, -0.1587, ?0.0409, ?0.0312],
? ? ? ? [ 0.1057, ?0.1348, -0.0590, -0.1538, ?0.2505, ?0.0651, -0.2461, -0.1856,
? ? ? ? ? 0.2498, -0.1969, ?0.0013, ?0.1979],
? ? ? ? [-0.1812, ?0.1153, ?0.2723, -0.2190, ?0.0371, -0.0341, ?0.2282, ?0.1461,
? ? ? ? ? 0.1890, ?0.1762, ?0.2657, -0.0827],
? ? ? ? [-0.0188, ?0.0081, -0.2674, -0.1858, ?0.1296, ?0.1728, -0.0770, ?0.1444,
? ? ? ? ?-0.2360, -0.1793, ?0.1921, -0.2791]], requires_grad=True)
fc1.bias
Parameter containing:
tensor([-0.0020, ?0.0985, ?0.1859, -0.0175], requires_grad=True)

代碼示例二

print(model.fc1.weight.requires_grad) ?# 可學(xué)習(xí)參數(shù)fc1.weight的requires_grad屬性

for name, param in model.named_parameters():
? ? if ("fc1" in name):
? ? ? ? param.requires_grad = False

print(model.fc1.weight.requires_grad) ?# 修改后可學(xué)習(xí)參數(shù)fc1.weight的requires_grad屬性

結(jié)果

(bbn) jyzhang@admin2-X10DAi:~/test$ python net.py
True
False

parameters()

總述

model.parameters()返回的是一個(gè)生成器,該生成器中只保存了可學(xué)習(xí)、可被優(yōu)化器更新的參數(shù)的具體的參數(shù),可通過循環(huán)迭代打印參數(shù)。(參見代碼示例一)

與model.named_parameters()相比,model.parameters()不會(huì)保存參數(shù)的名字。

該方法可以用來改變可學(xué)習(xí)、可被優(yōu)化器更新參數(shù)的requires_grad屬性,但由于其只有參數(shù),沒有對(duì)應(yīng)的參數(shù)名,所以當(dāng)要修改指定的某些層的requires_grad屬性時(shí),沒有model.named_parameters()方便。(參見

代碼示例二)

代碼示例一

# model.parameters()的用法
print(type(model.parameters()))

for param in model.parameters():
? ? print(param)

結(jié)果

(bbn) jyzhang@admin2-X10DAi:~/test$ python net.py
<class 'generator'>
Parameter containing:
tensor([1., 1., 1.], requires_grad=True)
Parameter containing:
tensor([0., 0., 0.], requires_grad=True)
Parameter containing:
tensor([[ 0.0036, ?0.1960, ?0.2315, -0.2408, ?0.1217, ?0.2579, -0.0676, -0.1880,
? ? ? ? ?-0.2855, -0.1587, ?0.0409, ?0.0312],
? ? ? ? [ 0.1057, ?0.1348, -0.0590, -0.1538, ?0.2505, ?0.0651, -0.2461, -0.1856,
? ? ? ? ? 0.2498, -0.1969, ?0.0013, ?0.1979],
? ? ? ? [-0.1812, ?0.1153, ?0.2723, -0.2190, ?0.0371, -0.0341, ?0.2282, ?0.1461,
? ? ? ? ? 0.1890, ?0.1762, ?0.2657, -0.0827],
? ? ? ? [-0.0188, ?0.0081, -0.2674, -0.1858, ?0.1296, ?0.1728, -0.0770, ?0.1444,
? ? ? ? ?-0.2360, -0.1793, ?0.1921, -0.2791]], requires_grad=True)
Parameter containing:
tensor([-0.0020, ?0.0985, ?0.1859, -0.0175], requires_grad=True)

代碼示例二

print(model.fc1.weight.requires_grad)

for param in model.parameters():
? ? param.requires_grad = False

print(model.fc1.weight.requires_grad)

結(jié)果

(bbn) jyzhang@admin2-X10DAi:~/test$ python net.py
True
False

state_dict()

總述

model.state_dict()返回的是一個(gè)有序字典OrderedDict,該有序字典中保存了模型所有參數(shù)的參數(shù)名和具體的參數(shù)值,所有參數(shù)包括可學(xué)習(xí)參數(shù)和不可學(xué)習(xí)參數(shù),可通過循環(huán)迭代打印參數(shù),因此,該方法可用于保存模型,當(dāng)保存模型時(shí),會(huì)將不可學(xué)習(xí)參數(shù)也存下,當(dāng)加載模型時(shí),也會(huì)將不可學(xué)習(xí)參數(shù)進(jìn)行賦值。(參見代碼示例一)

一般在使用model.state_dict()時(shí)會(huì)使用該函數(shù)的默認(rèn)參數(shù),model.state_dict()源碼如下:

# torch.nn.modules.module.py
class Module(object):
? ? def state_dict(self, destination=None, prefix='', keep_vars=False):
? ? ? ? if destination is None:
? ? ? ? ? ? destination = OrderedDict()
? ? ? ? ? ? destination._metadata = OrderedDict()
? ? ? ? destination._metadata[prefix[:-1]] = local_metadata = dict(version=self._version)
? ? ? ? for name, param in self._parameters.items():
? ? ? ? ? ? if param is not None:
? ? ? ? ? ? ? ? destination[prefix + name] = param if keep_vars else param.data
? ? ? ? for name, buf in self._buffers.items():
? ? ? ? ? ? if buf is not None:
? ? ? ? ? ? ? ? destination[prefix + name] = buf if keep_vars else buf.data
? ? ? ? for name, module in self._modules.items():
? ? ? ? ? ? if module is not None:
? ? ? ? ? ? ? ? module.state_dict(destination, prefix + name + '.', keep_vars=keep_vars)
? ? ? ? for hook in self._state_dict_hooks.values():
? ? ? ? ? ? hook_result = hook(self, destination, prefix, local_metadata)
? ? ? ? ? ? if hook_result is not None:
? ? ? ? ? ? ? ? destination = hook_result
? ? ? ? return destination

在默認(rèn)參數(shù)下,model.state_dict()保存參數(shù)時(shí)只會(huì)保存參數(shù)(Tensor對(duì)象)的data屬性,不會(huì)保存參數(shù)的requires_grad屬性,因此,其保存的參數(shù)的requires_grad的屬性變?yōu)镕alse,沒有辦法改變r(jià)equires_grad的屬性,所以改變r(jià)equires_grad的屬性只能通過上面的兩種方式。(參見代碼示例二)

model.state_dict()本質(zhì)上是淺拷貝,即返回的OrderedDict對(duì)象本身是新創(chuàng)建的對(duì)象,但其中的param參數(shù)的引用仍是模型參數(shù)的data屬性的地址,又因?yàn)門ensor是可變對(duì)象,因此,若對(duì)param參數(shù)進(jìn)行修改(在原地址變更數(shù)據(jù)內(nèi)容),會(huì)導(dǎo)致對(duì)應(yīng)的模型參數(shù)的改變。(參見代碼示例三)

代碼示例一

# model.state_dict()的用法
print(model.state_dict())

for name, param in model.state_dict().items():
? ? print(name)
? ? print(param)
? ? print(param.requires_grad)

結(jié)果

(bbn) jyzhang@admin2-X10DAi:~/test$ python net.py
OrderedDict([('bn1.weight', tensor([1., 1., 1.])), ('bn1.bias', tensor([0., 0., 0.])), ('bn1.running_mean', tensor([0., 0., 0.])), ('bn1.running_var', tensor([1., 1., 1.])), ('bn1.num_batches_tracked', tensor(0)), ('fc1.weight', tensor([[ 0.0036, ?0.1960, ?0.2315, -0.2408, ?0.1217, ?0.2579, -0.0676, -0.1880,
? ? ? ? ?-0.2855, -0.1587, ?0.0409, ?0.0312],
? ? ? ? [ 0.1057, ?0.1348, -0.0590, -0.1538, ?0.2505, ?0.0651, -0.2461, -0.1856,
? ? ? ? ? 0.2498, -0.1969, ?0.0013, ?0.1979],
? ? ? ? [-0.1812, ?0.1153, ?0.2723, -0.2190, ?0.0371, -0.0341, ?0.2282, ?0.1461,
? ? ? ? ? 0.1890, ?0.1762, ?0.2657, -0.0827],
? ? ? ? [-0.0188, ?0.0081, -0.2674, -0.1858, ?0.1296, ?0.1728, -0.0770, ?0.1444,
? ? ? ? ?-0.2360, -0.1793, ?0.1921, -0.2791]])), ('fc1.bias', tensor([-0.0020, ?0.0985, ?0.1859, -0.0175]))])
bn1.weight
tensor([1., 1., 1.])
False
bn1.bias
tensor([0., 0., 0.])
False
bn1.running_mean
tensor([0., 0., 0.])
False
bn1.running_var
tensor([1., 1., 1.])
False
bn1.num_batches_tracked
tensor(0)
False
fc1.weight
tensor([[ 0.0036, ?0.1960, ?0.2315, -0.2408, ?0.1217, ?0.2579, -0.0676, -0.1880,
? ? ? ? ?-0.2855, -0.1587, ?0.0409, ?0.0312],
? ? ? ? [ 0.1057, ?0.1348, -0.0590, -0.1538, ?0.2505, ?0.0651, -0.2461, -0.1856,
? ? ? ? ? 0.2498, -0.1969, ?0.0013, ?0.1979],
? ? ? ? [-0.1812, ?0.1153, ?0.2723, -0.2190, ?0.0371, -0.0341, ?0.2282, ?0.1461,
? ? ? ? ? 0.1890, ?0.1762, ?0.2657, -0.0827],
? ? ? ? [-0.0188, ?0.0081, -0.2674, -0.1858, ?0.1296, ?0.1728, -0.0770, ?0.1444,
? ? ? ? ?-0.2360, -0.1793, ?0.1921, -0.2791]])
False
fc1.bias
tensor([-0.0020, ?0.0985, ?0.1859, -0.0175])
False

代碼示例二

# model.state_dict()的用法
print(model.bn1.weight.requires_grad)
model.bn1.weight.requires_grad = False
print(model.bn1.weight.requires_grad)

for name, param in model.state_dict().items():
? ? if (name == "bn1.weight"):
? ? ? ? param.requires_grad = True

print(model.bn1.weight.requires_grad)

結(jié)果

(bbn) jyzhang@admin2-X10DAi:~/test$ python net.py
True
False
False

代碼示例三

# model.state_dict()的用法
print(model.bn1.weight)

for name, param in model.state_dict().items():
? ? if (name == "bn1.weight"):
? ? ? ? param[0] = 1000

print(model.bn1.weight)

結(jié)果

(bbn) jyzhang@admin2-X10DAi:~/test$ python net.py
Parameter containing:
tensor([1., 1., 1.], requires_grad=True)
Parameter containing:
tensor([1000., ? ?1., ? ?1.], requires_grad=True)

原文鏈接:https://blog.csdn.net/weixin_42149550/article/details/117128228

欄目分類
最近更新