Welcome to chainerboard’s documentation!

An unofficial visualization tool for chainer, inspired by tensorboard. The toolkit allows visualization of log from chainer.extensions.LogReport.

Example usage:

model = L.Classifier(MyModel())

optimizer = chainer.optimizers.Adam()
optimizer.setup(model)

train = create_my_data()
train_iter = chainer.iterators.SerialIterator(train, batchsize)

updater = training.StandardUpdater(train_iter, optimizer)
trainer = training.Trainer(updater, (epochs, 'epoch'), out='path/to/output')

trainer.extend(extensions.LogReport(log_name='my_log_data')))
# optional; allows visualization of parameters
trainer.extend(extensions.ParameterStatistics(model))

# Run the training
trainer.run()

and point chainerboard at the output log file to start local http server.

chainerboard path/to/output/my_log_name

now open http://localhost:6006/ to view the log.

Warning

The author of this project is not a professional web programmer. Never use the project on remote server since it may impose serious security risks.

Development

To setup development environment:

pip install -r requirements.txt

For testing,

tox

Build document

python setup.py build_sphinx

CHANGELOG

v0.1.4

  • Bugfix:

    • Auto-update was not working
    • It was occasionally causing OSError (#5)

v0.1.3

  • Bugfix:

    • Hot fix for a syntactic bug introduced in v0.1.2

v0.1.2

  • Bugfix:

    • Plotting fails when ‘Infinity’ is present (#2)

v0.1.1

  • Downgraded to more stable bootstrap 3
  • Connection error (distruption) is now handled more gracefully.
  • Prettified front end app.

v0.1.0

  • First alpha release.
  • Basic visualization of accuracy and loss.
  • Plot visualization using Plotly.js.
  • Visualization of histograms.
  • Parsing of log data from chainer’s LogReport
  • Log file watching using watchdog.
  • Real time updating using ajax (via angularJS).
  • Documentation using Sphinx.

Indices and tables