COMP6248-Reproducability-Challenge/MULTI-CLASS_CLASSIFICATION_WITHOUT_MULTI-CLASS_LABELS

Multi-class Classification without Multi-class Labels

Introduction

This repository provides reproduction of Multi-class Classification without Multi-class Labels

The open-source code can be found at: https://github.com/GT-RIPL/L2C

Team Member (Team Relu)

Ruihuang Ding (rd3n20@soton.ac.uk)

Bin Zhang (bz1u20@soton.ac.uk)

Yefan Zhuang (yz5e20@soton.ac.uk)

Requirement

PyTorch 1.0, 
python 2.7, 3.6, and 3.7
torch>=0.4.1
torchvision>=0.2.1
argparse
scipy
sklearn

Or you can just use the following command:

pip install -r requirements.txt

Our own implementation for evaluation

Supervised scenario

# The test files we used to test the MCL strategy

# generate the data in Table 1
python generate_table_test.py

# generate the accuracy and loss curve
python supervised_learning_compare_test.py

Unsupervised scenario

Learn the Similarity Prediction Network (SPN) with Omniglot_background and then transfer to the 20 alphabets in Omniglot_evaluation.

python demo_omniglot_transfer.py

Demo

Supervised Classification/Clustering with only pairwise similarity

# A quick trial:
python demo.py  # Default Dataset:MNIST, Network:LeNet, Loss:MCL
python demo.py --loss KCL

# Lookup available options:
python demo.py -h

# For more examples:
./scripts/exp_supervised_MCL_vs_KCL.sh

Unsupervised Clustering (Cross-task Transfer Learning)

# Learn the Similarity Prediction Network (SPN) with Omniglot_background and then transfer to the 20 alphabets in Omniglot_evaluation.
# Default loss is MCL with an unknown number of clusters (Set a large cluster number, i.e., k=100)
# It takes about half an hour to finish.
python demo_omniglot_transfer.py

# An example of using KCL and set k=gt_#cluster
python demo_omniglot_transfer.py --loss KCL --num_cluster -1

# Lookup available options:
python demo_omniglot_transfer.py -h

# Other examples:
./scripts/exp_unsupervised_transfer_Omniglot.sh