Tensorboard visualization

Because when doing experiments , Need to record data , For the convenience of recording data , have access to tensorboard Record the experimental results during the operation , To compare the results of that experiment is the best .
tensorboard Common functions are , Save the model network structure , Save the loss of the training set , Let's introduce tensorboard Several common functions of .

1.tensorboard Panel Introduction

This panel image comes from :https://blog.csdn.net/Zhangsan0219/article/details/121763073
I wrote my notes again with reference to this author , Easy to record .

2. Draw a discount statistical chart

• Draw multiple pictures separately

# -*- coding： utf-8 -*-
# @Author ：WenShuai Xiao
# @Time：2022-05-30
# @File：Tensorboard_line.py
import os.path

from torch.utils.tensorboard import SummaryWriter
import numpy as np
from tqdm import tqdm

if __name__=='__main__':
    print(' Beginning of the test ')
    '''
     Draw line 
     Start command  tensorboard --logdir=logs
     Designated port  tensorboard --logdir=logs --port=6007
    '''

    #  Create editor , Save log 
    # log_dir Save the path 
    logs_path=os.path.join("./logs","trian")
    if not os.path.exists(logs_path):
        print(' path does not exist ')
        os.makedirs(logs_path)

    writer=SummaryWriter(log_dir=logs_path)

    #  Simulate adding each epoch Of loss and Acc
    for n_iter in tqdm(range(100)):
        # writer.add_scalar Function to add a scalar value 
        # add_scalar(tags, scalar_value, global_stop)
        # tags: string,  Record the label of the object 
        # scalar_valud: float,  Recorded value 
        # global_step: int, x Shaft value 

        writer.add_scalar('Loss/Train',np.random.random(),n_iter)
        writer.add_scalar('Loss/Test',np.random.random(),n_iter)
        writer.add_scalar('Accuracy/train',np.random.random(),n_iter)
        writer.add_scalar('Accuracy/test',np.random.random(),n_iter)

    writer.close()
    print(' End of test ')

After running, it will be in logs File under which records are generated , Then find the path , Enter cmd-> enter , You can enter directly under this path dos command

Then enter the following code , Copy the last link to the browser and you can see the result

#  Activate the environment 
conda activate pytorch
#  Use tensorboard Open the path 
tensorboard --dirlogs=logs

• Put multiple pictures together

 # =====================================================
    '''
    writer.add_scalars(main_tag, tag_scalar_dict, global_step=None)
     This function is used to combine multiple scalar Add to the same diagram 
    main_tag: string,  Parent label 
    tag_scalar_dict: dict,  Will be multiple scalar Dictionary objects grouped together ,
         The dictionary's key Express tag, value For recorded scalar value 
    global_step: int,  The image x Axis 
    '''
    # ====================================================
    for i in tqdm(range(100),desc=' Multiple pictures together '):
        writer2.add_scalars(
            'run_14h',{
                'xsinx':i*np.sin(i/5),
                'xcosx':i*np.cos(i/5),
                'tanx':np.tan(i/5)
            },i
        )
    writer2.close()

Show results

The histogram behind , Continue to add notes when I need them

3. Draw model drawing

# -*- coding： utf-8 -*-
# @Author ：WenShuai Xiao
# @Time：2022-05-30
# @File：Tensorboard_model.py
from torch import ones
from torch.utils.tensorboard import SummaryWriter
from torch.nn import Linear,ReLU,Sequential
import os

if __name__=='__main__':
    logs_path = os.path.join("./logs", "model_graph")
    if not os.path.exists(logs_path):
        print(' path does not exist ')
        os.makedirs(logs_path)

    writer=SummaryWriter(logs_path)
    #  Write a model 
    model=Sequential(
        Linear(1,10),
        ReLU(),
        Linear(10,1)
    )
    data=ones(10,1)

    '''
    add_graph
     Draw the structure of the model 
     Double click the model to view 
    '''

    writer.add_graph(model=model,input_to_model=data)
    writer.close()

View the model diagram :

4. Multi model comparison

Sometimes it is necessary to adjust and compare the results of the experiment , This method can be used to save the results of each experiment to tensorboard in , If you learn bash grammar , Directly available bash Syntax automatically passes parameters , So you don't have to operate by yourself , Just wait patiently for the experimental results .

# -*- coding： utf-8 -*-
# @Author ：WenShuai Xiao
# @Time：2022-05-30
# @File：Tensorboard_mutil_model.py

import os
import datetime
from torch.utils.tensorboard import  SummaryWriter
if __name__=='__main__':
    #  Dynamic generation time 
    log_dir=os.path.join('logs',datetime.datetime.now().strftime("%Y%m%d_%H%M%S"))
    if not os.path.exists(log_dir):
        os.makedirs(log_dir)

    writer=SummaryWriter(log_dir=log_dir)
    # y=2x
    for i in range(100):
        writer.add_scalar(tag='y=2x',scalar_value=1.5*i,global_step=i)

    writer.close()

View experiment results :