tfaip 1.0.1

No project description provided

tf2_aip_base

This repository is designed as a research framework for supervised machine learning. It aims to reduce your work on the train-loop, validation, saving, optimizer, multi-gpu and provides lot more features, which can be configured via command line.

Setup

see: Setup

Usage

To setup your own scenario see: Scenario setup

Running the tutorial scenario:

The default tutorial scenario is 'fashion-mnist'. Run your first training with:

tfaip-train tutorial --trainer_params checkpoint_dir=models/fashion_default

tfaip-train refers to tfaip/scripts/train.py. You can switch to mnist data with --data_params dataset=mnist.

You can evaluate the model on the validation dataset with: tfaip-lav --export_dir models/fashion_default/export

Most hyper parameter can be configured via command line see tfaip-train -h and tfaip-train tutorial -h. Checkout the Wiki for further explanations.

Contributions are welcome, and they are greatly appreciated!