torch.manual_seed()_torch8项怎么看结果「建议收藏」

import torch
 
torch.manual_seed(0)
print(torch.randn(1, 2))
print(torch.randn(1, 2))

print("======================") # torch.manual_seed()控制torch中随机数的生成，并不控制numpy和random中
torch.manual_seed(0)
print(torch.randn(1, 2))
print(torch.randn(1, 2))

然后就有一个好玩的事情发生了

可以看到，我仅仅多设置了一个模型，导致这个随机种子不同了，这个原因到底在哪呢？我之前的测试代码在这

import torch.nn as nn

### 这里是最简单的simple autoencoder
def init_weights(m):
    """ initialize weights of fully connected layer """
    if type(m) == nn.Linear:
        nn.init.xavier_uniform_(m.weight)
        m.bias.data.fill_(0.01)

# Encoder
class Encoder(nn.Module):
    def __init__(self, num_inputs):
        super(Encoder, self).__init__()
        self.encoder = nn.Sequential(
            nn.Linear(num_inputs, 64),
            nn.ReLU(),
            nn.Linear(64,32))
        self.encoder.apply(init_weights)
    def forward(self, x):
        x = self.encoder(x)
        return x
# Decoder
class Decoder(nn.Module):
    def __init__(self, num_inputs):
        super(Decoder, self).__init__()
        self.decoder = nn.Sequential(
            nn.Linear(32, 64),
            nn.ReLU(),
            nn.Linear(64, num_inputs),
           	nn.ReLU())
        self.decoder.apply(init_weights)
    def forward(self, x):
        x = self.decoder(x)
        return x
# Autoencoder
class autoencoder(nn.Module):
    def __init__(self, num_inputs):
        super(autoencoder, self).__init__()
        self.encoder = Encoder(num_inputs)
        self.decoder = Decoder(num_inputs)
    def forward(self, x):
        code = self.encoder(x)
        x = self.decoder(code)
        return code, x

不建立autoencoder，看dataloader的数据，第一次运行

from torch.utils.data import TensorDataset,DataLoader
import torch
from sklearn.datasets import load_iris
import random
import os 
import numpy as np 
def seed_torch(seed=42):
    random.seed(seed)
    os.environ['PYTHONHASHSEED'] = str(seed)
    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.badatahmark = False
    torch.backends.cudnn.deterministic = True
    torch.backends.cudnn.enabled = False
    
seed_torch(1)
X,y=load_iris(return_X_y=True)
myset=TensorDataset(torch.FloatTensor(X),torch.FloatTensor(y))
myloader=DataLoader(myset,batch_size=10,shuffle=True,num_workers=0)

for idx, (X,y) in enumerate(myloader):
    print(X)
    print(y)
    break
print("=============================================================================================")
#model=autoencoder(1000)
for idx, (X,y) in enumerate(myloader):
    print(X)
    print(y)
    break

当再次执行这个代码，结果还是固定的

不建立autoencoder，看dataloader的数据，第二次运行

from torch.utils.data import TensorDataset,DataLoader
import torch
from sklearn.datasets import load_iris
import random
import os 
import numpy as np 
def seed_torch(seed=42):
    random.seed(seed)
    os.environ['PYTHONHASHSEED'] = str(seed)
    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.badatahmark = False
    torch.backends.cudnn.deterministic = True
    torch.backends.cudnn.enabled = False
    
seed_torch(1)
X,y=load_iris(return_X_y=True)
myset=TensorDataset(torch.FloatTensor(X),torch.FloatTensor(y))
myloader=DataLoader(myset,batch_size=10,shuffle=True,num_workers=0)

for idx, (X,y) in enumerate(myloader):
    print(X)
    print(y)
    break
print("=============================================================================================")
#model=autoencoder(1000)
for idx, (X,y) in enumerate(myloader):
    print(X)
    print(y)
    break

目前这里还没有问题，一切都如所料

建立autoencoder，看dataloader的数据，第一次运行

from torch.utils.data import TensorDataset,DataLoader
import torch
from sklearn.datasets import load_iris
import random
import os 
import numpy as np 
def seed_torch(seed=42):
    random.seed(seed)
    os.environ['PYTHONHASHSEED'] = str(seed)
    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.badatahmark = False
    torch.backends.cudnn.deterministic = True
    torch.backends.cudnn.enabled = False
    
seed_torch(1)
X,y=load_iris(return_X_y=True)
myset=TensorDataset(torch.FloatTensor(X),torch.FloatTensor(y))
myloader=DataLoader(myset,batch_size=10,shuffle=True,num_workers=0)

for idx, (X,y) in enumerate(myloader):
    print(X)
    print(y)
    break
print("=============================================================================================")
model=autoencoder(1000)
for idx, (X,y) in enumerate(myloader):
    print(X)
    print(y)
    break

这个时候，结果如下
可以看到，这个时候就不对劲了，我加入了一个autoencoder的模型，可以看到，之后生成的dataloader就开始变化了，不再是以前的顺序了

建立autoencoder，看dataloader的数据，第二次运行

from torch.utils.data import TensorDataset,DataLoader
import torch
from sklearn.datasets import load_iris
import random
import os 
import numpy as np 
def seed_torch(seed=42):
    random.seed(seed)
    os.environ['PYTHONHASHSEED'] = str(seed)
    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.badatahmark = False
    torch.backends.cudnn.deterministic = True
    torch.backends.cudnn.enabled = False
    
seed_torch(1)
X,y=load_iris(return_X_y=True)
myset=TensorDataset(torch.FloatTensor(X),torch.FloatTensor(y))
myloader=DataLoader(myset,batch_size=10,shuffle=True,num_workers=0)

for idx, (X,y) in enumerate(myloader):
    print(X)
    print(y)
    break
print("=============================================================================================")
model=autoencoder(1000)
#torch.manual_seed("1")
for idx, (X,y) in enumerate(myloader):
    print(X)
    print(y)
    break

for idx, (X,y) in enumerate(myloader):
    print(X)
    print(y)
    break

虽然这个结果与加入autoencoder后第一次运行的结果相同，但是这个结果还是与没有加入autoencoder前的生成数据不同了

问题所在

你可以理解 torch.manual_seed()只能用一次，你最开始设置的随机种子在没有加入模型前测试中，作用在了dataloader上，而加入模型后，你的随机种子首先作用于设置模型的权重，然后再是dataloader,这个顺序也就变了，所以要想获得建立模型后，还与之前一样的dataloader的抽样结果，就在dataloaer之前重新设置torch.manual_seed()

from torch.utils.data import TensorDataset,DataLoader
import torch
from sklearn.datasets import load_iris
import random
import os 
import numpy as np 
def seed_torch(seed=42):
    random.seed(seed)
    os.environ['PYTHONHASHSEED'] = str(seed)
    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.badatahmark = False
    torch.backends.cudnn.deterministic = True
    torch.backends.cudnn.enabled = False
    
seed_torch(1)
X,y=load_iris(return_X_y=True)
myset=TensorDataset(torch.FloatTensor(X),torch.FloatTensor(y))
myloader=DataLoader(myset,batch_size=10,shuffle=True,num_workers=0)

# for idx, (X,y) in enumerate(myloader):
# print(X)
# print(y)
# break
print("=============================================================================================")
model=autoencoder(1000)
torch.manual_seed("1")
for idx, (X,y) in enumerate(myloader):
    print(X)
    print(y)
    break

for idx, (X,y) in enumerate(myloader):
    print(X)
    print(y)
    break

可以看到这个结果与之前没有加入autoencoder的结果是一致的，是不是很神奇

今天的文章torch.manual_seed()_torch8项怎么看结果「建议收藏」分享到此就结束了，感谢您的阅读。

版权声明：本文内容由互联网用户自发贡献，该文观点仅代表作者本人。本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如发现本站有涉嫌侵权/违法违规的内容，请发送邮件至举报，一经查实，本站将立刻删除。
如需转载请保留出处：https://bianchenghao.cn/86313.html