Torch tensorboard
Collection of snippets of tensorboard usage with pytorch using the tensorboardX library.
Requirements
pip install torch
pip install tensorflow
pip install tensorboard
Run tensorboard
tensorboard --logdir runs
Open up the webapp in your browser, usually at http://localhost:6006.
Training script
import numpy as np
from torch.utils.tensorboard import SummaryWriter
## clean run folder
import shutil
import time
shutil.rmtree('runs', ignore_errors=True)
## run 1
writer = SummaryWriter(flush_secs=1)
for x in range(100):
noise = np.random.normal(scale=1)
y = (x/20)**2 + noise
writer.add_scalar('noisy_parabola', y, x)
time.sleep(0.1)
writer.close()
## run 2
writer = SummaryWriter(flush_secs=1)
for x in range(100):
noise = np.random.normal(scale=6)
y = (x/20)**2 + noise
writer.add_scalar('noisy_parabola', y, x)
time.sleep(0.1)
writer.close()
The tensorboard plot will be updated in real time. The final output will be something similar to this:
