https://github.com/maxme1/tensorboard-easy

A logger that writes in a format, readable by tensorboard
https://github.com/maxme1/tensorboard-easy

Last synced: 3 months ago
JSON representation

A logger that writes in a format, readable by tensorboard

• Host: GitHub
• URL: https://github.com/maxme1/tensorboard-easy
• Owner: maxme1
• License: mit
• Created: 2017-10-04T12:17:04.000Z (over 7 years ago)
• Default Branch: master
• Last Pushed: 2017-10-16T19:11:37.000Z (over 7 years ago)
• Last Synced: 2024-10-02T16:07:51.476Z (4 months ago)
• Language: Python
• Size: 35.2 KB
• Stars: 5
• Watchers: 1
• Forks: 3
• Open Issues: 1
• Metadata Files:
  • Readme: README.rst
  • License: LICENSE

Awesome Lists containing this project

README

        
A small logger that lets you write logs readable by Tensorboard but

doesn’t require Tensorflow.

Usage

=====

You can use the logger as a context manager:

.. code:: python

    from tensorboard_easy import Logger

    import numpy as np

    with Logger('/path/to/logs/folder/') as log:

        log.log_scalar('my_scalar', 100, step=1)

        log.log_image('my_images', np.random.rand(3, 20, 20), step=1)

or you can close the logger explicitly:

.. code:: python

    log = Logger('/some/other/logs')

    log.log_text('my_text', "Let's throw in some text", 0)

    log.log_text('my_text', [['Some', 'tensor'], ['with', 'text!']], 1)

    log.log_histogram('my_histogram', np.random.rand(500), step=0)

    log.close()

It supports scalars, images, text and histograms.

You can also create functions, that write to a specific tag and automatically

increase the step:

.. code:: python

    with Logger('/path/to/logs/folder/') as log:

        write_loss = log.make_log_scalar('loss')

        for i in range(1, 100):

            write_loss(1 / i)

Installation

============

It can be installed via pip:

``pip install tensorboard-easy``

The ``tensorflow`` or ``tensorflow-tensorboard`` packages are not

required, however you will need one of them to visualize your logs.