https://github.com/saravanabalagi/keras_tensorboard_starter

Starter project for Keras with Tensorboard logging
https://github.com/saravanabalagi/keras_tensorboard_starter

keras tensorboard tensorflow tensorflow2

Last synced: 3 months ago
JSON representation

Starter project for Keras with Tensorboard logging

• Host: GitHub
• URL: https://github.com/saravanabalagi/keras_tensorboard_starter
• Owner: saravanabalagi
• License: mit
• Created: 2020-04-16T21:32:54.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2020-04-17T22:34:54.000Z (about 6 years ago)
• Last Synced: 2025-03-27T14:53:21.747Z (over 1 year ago)
• Topics: keras, tensorboard, tensorflow, tensorflow2
• Language: Python
• Homepage:
• Size: 81.1 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Keras Tensorboard Starter

This is a keras starter project that provides tf.summary logging out of the box 

inside your model inference code for both train and eval runs.

You should be able to see something like

![Screenshot](images/screenshot.png)

where 

- eval has exactly 3 records (at 0, 10 and 20 corresponding to 3 `evaluate`s)

- train has 20 records (0-9 and 10-19 corresponding to 2 `fit`s with 10 epochs each)

 

## Instructions

 

From Tensorflow 2.2, we can override Keras `model.fit` using `Model.train_step`. 

This allows us to write a custom train step and use the same old `model.fit` in an elegant way, 

but to call our custom train step logic. See example at [train_and_evaluate.py](train_and_evaluate.py)

## For Older Versions

For older versions of tensorflow, the hack implemented in [train_and_evaluate_old.py](train_and_evaluate_old.py) 

where we can pass the summary writer into an argument

and use it in `call` method to log can be a workaround.

Although it's not as elegant as the above method, it works!

![Screenshot Old](images/screenshot_old.png)

## Licence

Please see attached [Licence](LICENCE)