https://github.com/makseq/tfmicro
Micro framework for Tensorflow
https://github.com/makseq/tfmicro
micro-framework model tensorflow train
Last synced: 7 months ago
JSON representation
Micro framework for Tensorflow
- Host: GitHub
- URL: https://github.com/makseq/tfmicro
- Owner: makseq
- Created: 2018-08-28T12:40:26.000Z (about 7 years ago)
- Default Branch: master
- Last Pushed: 2020-10-16T12:08:24.000Z (almost 5 years ago)
- Last Synced: 2025-01-25T09:25:16.354Z (9 months ago)
- Topics: micro-framework, model, tensorflow, train
- Language: Python
- Size: 140 KB
- Stars: 3
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# TfMicro
TfMicro is a micro framework for Tensorflow training similar to Keras but easy and native to Tensorflow.
Focus only on creating your TF model. All the rest TfMicro will be done for you.
Multiprocessing data template, callbacks, keyboard operations, saving/loading/preloading models and much more.
# Features
* Multiprocessing data template: inherit from tfmicro.Data class to feed your data into model with multiprocessing.
* Callbacks
* ModelCheckpoint: to save models by monitor value
* Keyboard stop: to stop training by pressing 'q'
* Validation/KeyboardValidation: to make validation step when you want
* ReducingLearningRate/KeyboardLearningRate: change learning rate on the fly by pressing '+'/'-' keys.
* Testarium: connect Testarium with tfmicro together
* AccuracyCallback: evaluate accuracy by model.logits & model.labels
* FafrCallback: evaluate DET curve and EER on epoch end
* Progress bar and info while training
* Custom user indicators and progress bars
* Custom user keyboard operations: bind any key to any actions
* Saving/loading/preloading models and weights
* Tensorboard automatic support
* Production code templates: training and inference are strong separated
# Install
```
git clone git@github.com:makseq/tfmicro.git
cd tfmicro
python setup.py develop
```
# Usage
See example folder.