DTS

Overview

This is the code of our Neural Networks submission.

Environment Setup

To replicate our environment and ensure seamless execution of the code, please rebuild the environment using the provided environment.yaml file. You can do this with the following Conda command:

conda env create -f environment.yaml

Usage

CIFAR-100

Pretrain

python pretrain.py --dataset CIFAR100 --n_labels 5000 --n_unlabels 20000 --n_valid 10000 --n_class 50 --ratio 0.6 --warm_up 200

Run

CIFAR-100 with class mismatch ratio 0.3

python train.py --dataset CIFAR100 --n_labels 5000 --n_unlabels 20000 --n_valid 10000 --n_class 50 --ratio 0.3 --lm 0.5 --threshold 0.85 --Ctk_weight 0.25 --Ctu_weight 0.1 --socr_s2_weight 0.3

CIFAR-100 with class mismatch ratio 0.6

python train.py --dataset CIFAR100 --n_labels 5000 --n_unlabels 20000 --n_valid 10000 --n_class 50 --ratio 0.6 --lm 0.5 --threshold 0.85 --Ctk_weight 0.25 --Ctu_weight 0.1 --socr_s2_weight 0.3