Tensorflow 2.0 Project Template

A tensorflow 2.0 project template that is redesigned to automate model creation using Keras (more to be added later) for competitions. Thus Faster creation, training and evaluation of models.

template files are provided with an example and sample configs, data files from the titanic kaggle competitions were used

Acknowledgments

Thanks to Mahmoud Gemy for developing the template. The template is a combination of the templates provided here: https://github.com/MrGemy95/Tensorflow-Project-Template, Mahmoud Gemy https://github.com/Ahmkel/Keras-Project-Template , Ahmed Hamada Mohamed Kamel El-Hinidy with added improvements and automation for faster training and evaluating models based on configs provided.

Folder structure

├──  base
│   ├── base_model.py   - this file contains the abstract class of the model.
│   └── base_data_loader.py   - this file contains the abstract class of the data loader.
│   └── base_trainer.py   - this file contains the abstract class of the trainer.
│
│
├── models              - this folder contains the models for the project.
│   └── model_01.py
│
│
├── trainers             - this folder contains trainers of your project.
│   └── trainer.py
│   
│  
├──  data _loader  
│    └── data_loader_01.py  - data loader responsible for handling data generation and preprocessing
│
│
├── train.py --  main used to run the training across different config files and models
│
├── evaluate.py --  files responsible for the evaluation of different models. Loading and selecting the best model
│ 
├── train_bash.sh --  example bash script to run the training with different arguments
│
└── utils
     ├── dirs.py
     └── factory.py
     └── config.py
     └── utils.py

Configuration

Config files are used to choose the models, trainers, and dataloader files for each project. so multiple project can co-exist in this template. Check the config file contents for more info.

Future Work

• Add tensorflow dataset API and feature columns [ some lines added, not complete due to tensorflow bug]
• randomized config file generator
• Kaggle submission availability
• More examples on image and text data

More Info

For further info on how the template is built and more about its core components please check the link in the Acknowledgments