BuilT(Build a Trainer)

Easily build a trainer for your Depp Neural Network model and experiment as many as you want to find optimal combination of components(model, optimizer, scheduler) and hyper-parameters in a well-organized manner.

• No more boilerplate code to train and evaluate your DNN model. just focus on your model.
• Simply swap your dataset, model, optimizer and scheduler in the configuration file to find optimal combination. Your code doesn't need to be changed!!!.
• Support Cross Validation, OOF(Out of Fold) Prediction
• Support WandB(https://wandb.ai/) or tensorboard logging.
• Support checkpoint management(Save and load a model. Resume the previous training)
• BuilT easily integrates with Kaggle(https://www.kaggle.com/) notebook. (todo: add notebook link)

Installation

Please follow the instruction below to install BuilT.

Installation of BuilT package from the source code

git clone https://github.com/UoA-CARES/BuilT.git
cd BuilT
python setup.py install

Installation of BuilT package using pip

BuilT can be installed using pip(https://pypi.org/project/BuilT/).

pip install built

Usage

Configuration

Builder

Trainer

Dataset

Model

Loss

Optimizer

Scheduler

Logger

Metric

Inference

Ensemble

Examples

MNIST hand-written image classification

(todo)

Sentiment Classification

(todo)

Developer Guide

(todo)

conda create -n conda_BuilT python=3.7
conda activate conda_BuilT
pip install -r requirements.txt

Reference

https://packaging.python.org/tutorials/packaging-projects/