Regression-Datasets

This repository offers a diverse collection of regression datasets across vision, audio and text domains. It provides dataset classes that follow the PyTorch Datasets structure, allowing users to automatically download and load these datasets with ease. All datasets come with a permissive license, permitting their use for research purposes.

1. Installation

To install the regsets package, you can use pip:

python -m pip install regsets

Alternatively, you can download a specific dataset file (e.g., utkface.py) and include it in your project to load the dataset locally.

2. Usage

Below are examples of how to use the regsets package for loading datasets.

Vision Datasets

from regsets.vision import UTKFace

utkface_trainset = UTKFace(root="./data", split="train", download=True)

for image, label in utkface_trainset:
    ...

Audio Datasets

from regsets.audio import VCC2018

vcc2018_trainset = VCC2018(root="./data", split="train", download=True)

for audio, sample_rate, label in vcc2018_trainset:
    ...

Text Datasets

from regsets.text import Amazon_Review

amazon_review_trainset = Amazon_Review(root="./data", split="train", download=True)

for texts, label in amazon_review_trainset:
    (ori, aug_0, aug_1) = texts
    ...

3. Datasets

For datasets that do not provide a predefined train-test split, I randomly sample 80% of the data for training and reserve the remaining 20% for testing. Details for each dataset are provided below.

Vision Datasets

Audio Datasets

Text Datasets

4. License

Distributed under the MIT License. See LICENSE for more information.

5. Contact

• Pin-Yen Huang ([email protected])

6. Acknowledgments

• PyTorch
• UTKFace
• VCC2018
• BVCC
• USB
• Amazon Review
• Yelp Review
• README Template