2.2 KiB
Story Point Estimator
This repository contains code for training and evaluating and exporting models for story point estimation.
The dataset used for training is the IEEE TSE2018 dataset which contains user stories from 16 open-source projects from 9 different organizations.
Currently, the following models are supported:
Setup
Create and activate conda environment
conda create --name spestimator python=3.8
conda activate spestimator
Install dependencies
pip install torch==2.4.1 --index-url https://download.pytorch.org/whl/cu124
pip install -r requirements.txt
Setup dotenv
Create a .env file in the root directory and add the following environment variables:
DATA_PATH=<path to data directory>
Usage
Train models for story point estimation
See training options
python train.py --help
Train a model with your desired options while also seing validation results
python train.py <options>
Users can see in the experiments folder the results inside a subfolder with their experiment's name. The folder contains:
- a checkpoints folder where the results of each epoch and their according performance metrics are stored
- a params.json file with the experiment's parameters
- a log file with the experiment's logs regarding the training process
Evaluate existing models
See evaluation options
python eval.py --help
Evaluate a model with your desired options
python eval.py <options>
Export models
See export options
python export.py --help
Export a model with your desired options
python export.py <options>
Exports the model and tokenizer to the specified directory. The model is stored in a .safetensors file (for python users) and an .onnx file for other languages, and the tokenizer is stored in a vocab.txt file and 4 json files: config.json, special_tokens_map.json, tokenizer_config.json, and tokenizer.json.