Basic Tensorboard with Code

Tensorboard

• Supports PyTorch
• Shows scalars (accuracy, loss etc)
• Shows the computational graph
• Shows the histogram of weights

1. How to Use
   • Create directory to save log

     import os
     log_dir = 'log_file'
     os.makedirs(log_dir)
     

   • Import tensorboard log writer

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

   • create different experiments

     experiment_dir = log_dir + '/exp1'
     writer = SummaryWriter(experiment_dir)
     

   • Log content

     for iter_nth in range(100):
      writer.add_scalar('loss/train' value, iter_nth)
     writer.flush() # save log on the drive
     

   • Activate Tensorboard on Jupyter Notebook

     %load_ext tensorboard
     %tensorboard --logdir {log_dir}
     

   • Use Tensorboard

     tensorboard --logdir PATH --host ADDR --port PORT