This is an implementation on Python 3.8 to analyze the performance differences among various algorithms involved in dimensionality reduction (DR), namely t-SNE, UMAP, PaCMAP and h-NNE.
This code is submitted as a course project of the PHD course COMP9102-2022 at HKUCS.
You would require the following packages to fully use this code on your machine according to installation file:
- scikit-learn
- numpy
- numba
- torchvision
- umap-learn
- pacmap
- hnne
- annoy
- tqdm
- p_tqdm
You can use pip to install the above packages from PyPI. It will automatically install the dependencies for you:
pip install scikit-learn==0.24.1
pip install numpy==1.19
pip install numba==0.51.0
pip install torchvision==0.2.1
pip install -U umap-learn==0.5.1
pip install pacmap==0.6.3
pip install hnne
pip install annoy
pip install -U tqdm==4.45.0
pip install -U p_tqdm==1.2
To evaluate the differences among various DR performance on three standard datasets, namely MNIST, SVHN and CIFAR-10, you can run a command as follows,
python evaluation.py
After the code running, you can obtain a csv file, where you can find the results of KNN Accuracy, SVM Accuracy, Random Triplet Accuracy, Centroid Triplet Accuracy and running time.
To visualize the features obtained from diverse DR algorithms as stated above, you can run a command as follows,
python visualization.py
Thanks to PaCMAP. We obtain great inspiration and reference from this code to implement our work.