当前位置：网站首页>Using pytorch's tensorboard visual deep learning indicators | pytorch series (25)

Using pytorch's tensorboard visual deep learning indicators | pytorch series (25)

2022-07-28 02:53:00 【51CTO】

writing |AI_study

Use PyTorch Of TensorBoard- Visualizing deep learning indicators | PyTorch series （ twenty-five ）_ The Internet

The original title ：TensorBoard With PyTorch - Visualize Deep Learning Metrics

At this point in the series , We have just completed the network operation during the training . Now? , We hope to get more indicators about this process , To better understand the situation behind the scenes .

A bird's-eye view of where we are in the training process .

Prepare the data
Build a model
Training models
Analysis of the results of the model

Use for this TensorBoard

TensorBoard：TensorFlow Visual toolkit for

TensorBoard It provides the visualization and tools for machine learning experiments ：

Tracking and visualization indicators , Such as loss and accuracy
Visual model diagram （ Operation and layer ）
See the weight , Histograms of deviations or other tensors over time
take embedding Project to a lower dimensional space
Display images , Text and audio data
analysis TensorFlow Program
And more

from PyTorch 1.1.0 version ,PyTorch Added a tensorboard Utility packages , So that we can TensorBoard And PyTorch Use it together .

       
       print(
       
       torch
       
       .__version__)
       
       1.1.0
      
1.
2.

       
       from 
       
       torch.
       
       utils.
       
       tensorboard 
       
       import 
       
       SummaryWriter
      
1.

by PyTorch install TensorBoard

for PyTorch install TensorBoard, Please perform the following steps ：

Confirm that you are running PyTorch edition 1.1.0 Or later .
Confirm that you are running TensorBoard 1.15 Or later .
Please note that ,PyTorch The use of TensorBoard And for TensorFlow Created TensorBoard identical . Use the following command to check the TensorBoard Version of ：

       
       tensorboard --version
      
1.

4. Use the following command to install TensorBoard.

       
       pip install 
       
       tensorboard
      
1.

Once installed TensorBoard 1.15 Or later , We can start ！

Please note that ,PyTorch Document theory TensorBoard edition 1.14 It's necessary .

PyTorch Of TensorBoard introduction

TensorBoard It's at the end of a font Web Interface , Actually read the data from the file and display it . To use TensorBoard, Our task is to save the data we want to display to TensorBoard Can read the file .

In order to make us more relaxed ,PyTorch Created a new one called SummaryWriter Utility classes for . To access this class , We use the following import ：

       
       from 
       
       torch.
       
       utils.
       
       tensorboard 
       
       import 
       
       SummaryWriter
      
1.

After importing this class , We can create instances of this class , It is then used to get data from the program to the file system , Then from TensorBoard To use .

Network diagram and training set image

SummaryWriter Class has many methods , We can call these methods to selectively select and select what we want TensorBoard Available data . First , We deliver the network and a batch of images to the writers .

       
       tb 
       
       = SummaryWriter
       
       (
       
       )
       
network 
       
       = Network
       
       (
       
       )
       
images, labels 
       
       = next
       
       (iter
       
       (train_loader
       
       )
       
       )
       
grid = torchvision.utils
       
       .make_grid
       
       (images
       
       )
       
tb
       
       .add_image
       
       (
       
       'images', grid
       
       )
       
tb
       
       .add_graph
       
       (network, images
       
       )
       
tb
       
       .close
       
       (
       
       )
      
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.

The code for TensorBoard Create a file called tb Of SummaryWriter example . then , Create our PyTorch Examples of the Internet , And from our PyTorch Decompress a batch of images and labels in the data loader object .

then , Add images and networks to TensorBoard In the file to be used . actually , It can be said that network diagram and image batch Have been added to TensorBoard in .

‍

function TensorBoard

To start the TensorBoard, We need to run on the terminal tensorboard command . This will start a local server , The server will be TensorBoard UI And our SummaryWriter Data written to disk provides services .

By default ,PyTorch SummaryWriter Object writes data to a file named ./runs Disk in the directory of .

When running tensorboard On command , We pass a parameter to tell tensorboard Where is the data . Just like this. ：

       
       tensorboard --logdir=
       
       runs
      
1.

TensorBoard The server will start and is listening on port 6006 Upper http request . These details will be displayed in the console .

Access by browsing to TensorBoard UI：

       
       http:
       
       //localhost:6006
      
1.

ad locum , We will be able to see our network diagram and image data . at present , This really provides us with visual effects , But not as useful as the next .

TensorBoard Histogram and scalar

We can add to TensorBoard The next data import type of is digital data . We can add over time or epoch Scalar value displayed . We can also add values to the histogram to see the frequency distribution of values .

To add scalars and histograms , We use PyTorch SummaryWriter Class .

This is an example of a call ：

       
       tb
       
       .
       
       add_scalar(
       
       'Loss', 
       
       total_loss, 
       
       epoch)
       
       tb
       
       .
       
       add_scalar(
       
       'Number Correct', 
       
       total_correct, 
       
       epoch)
       
       tb
       
       .
       
       add_scalar(
       
       'Accuracy', 
       
       total_correct 
       
       / 
       
       len(
       
       train_set), 
       
       epoch)
       
       tb
       
       .
       
       add_histogram(
       
       'conv1.bias', 
       
       network
       
       .
       
       conv1
       
       .
       
       bias, 
       
       epoch)
       
       tb
       
       .
       
       add_histogram(
       
       'conv1.weight', 
       
       network
       
       .
       
       conv1
       
       .
       
       weight, 
       
       epoch)
       
       tb
       
       .
       
       add_histogram(
       
       'conv1.weight.grad', 
       
       network
       
       .
       
       conv1
       
       .
       
       weight
       
       .
       
       grad, 
       
       epoch)
      
1.
2.
3.
4.
5.
6.
7.
8.

This is an example , It shows how to put these calls in the training loop ：

       
       network 
       
       = Network()
       
train_loader 
       
       = torch.utils.data.DataLoader(train_set, 
       
       batch_size
       
       =
       
       100)
       
optimizer 
       
       = optim.Adam(network.parameters(), 
       
       lr
       
       =
       
       0.01)
       
images, labels 
       
       = next(iter(train_loader))
       
grid 
       
       = torchvision.utils.make_grid(images)
       
tb 
       
       = SummaryWriter()
       
tb.add_image(
       
       'images', grid)
       
tb.add_graph(network, images)
       
       for epoch 
       
       in range(1):
       
    total_loss 
       
       = 
       
       0
       
    total_correct 
       
       = 
       
       0
       
       for batch 
       
       in train_loader: 
       
       # Get Batch
       
       # Pass Batch
       
       # Calculate Loss
       
       # Calculate Gradient
       
       # Update Weights
       
    tb.add_scalar(
       
       'Loss', total_loss, epoch)
       
    tb.add_scalar(
       
       'Number Correct', total_correct, epoch)
       
    tb.add_scalar(
       
       'Accuracy', total_correct / len(train_set), epoch)
       
    tb.add_histogram(
       
       'conv1.bias', network.conv1.bias, epoch)
       
    tb.add_histogram(
       
       'conv1.weight', network.conv1.weight, epoch)
       
    tb.add_histogram(
       
       'conv1.weight.grad'
       
        ,network.conv1.weight.grad
       
        ,epoch
       
    )
       
    print(
       
       "epoch", epoch, 
       
       "total_correct:", total_correct, 
       
       "loss:", total_loss
       
    )
       
tb.close()
      
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.

This adds these values to TensorBoard in . These values are even updated in real time during network training .

It is helpful to look at the loss and accuracy values over time . however , We may need to admit , It doesn't really need to be TensorBoard.

TensorBoard The real function of is that it has many functions of running out of the box . This enables us to quickly conduct experiments by changing the hyperparametric values and comparing runs , To see which parameters are most effective .

The content of the article is carefully studied , My level is limited , Translation cannot be perfect , But it really took a lot of effort , I hope you can move your sexy hands , Share a circle of friends , Support me ^_^

The original English link is ：

https://deeplizard.com/learn/video/pSexXMdruFM

Use PyTorch Of TensorBoard- Visualizing deep learning indicators | PyTorch series （ twenty-five ）_python_02

Use PyTorch Of TensorBoard- Visualizing deep learning indicators | PyTorch series （ twenty-five ）_ visualization _03

Use PyTorch Of TensorBoard- Visualizing deep learning indicators | PyTorch series （ twenty-five ）_ The Internet _04

Use PyTorch Of TensorBoard- Visualizing deep learning indicators | PyTorch series （ twenty-five ）_ visualization _05

原网站

版权声明
本文为[51CTO]所创，转载请带上原文链接，感谢
https://yzsam.com/2022/209/202207280142557735.html

当前位置：网站首页>Using pytorch's tensorboard visual deep learning indicators | pytorch series (25)

Using pytorch's tensorboard visual deep learning indicators | pytorch series (25)

边栏推荐

猜你喜欢

随机推荐