DMT-EV: An Explainable Deep Network for Dimension Reduction

The code includes the following modules:

• Training
• Inference
• Comparison with t-SNE, UMAP and PCA

Requirements

• torch>=2.3.1
• torchaudio>=2.3.1
• torchvision>=0.18.1
• pytorch-lightning==2.4.0

Installation

Create a new conda environment and install torch, torchvision, torchaudio:

conda create -n DMT python=3.10
conda activate DMT
pip install torch==2.3.1 torchvision==0.18.1 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu121

Then you can install the package from source or from PyPI. Install from source:

pip install -e git+https://github.com/Westlake-AI/DMT-learn.git#egg=dmt-learn

Install from PyPI:

pip install dmt-learn

Running the code

Use the following code to fit the model to the dataset and visualize the results.

import numpy as np
import matplotlib.pyplot as plt
from sklearn.datasets import load_iris
from dmt import DMT

# Load sample dataset
iris = load_iris()
X = iris.data
y = iris.target

# Perform t-SNE
dmt = DMT(num_fea_aim=100)
X_dmt = dmt.fit_transform(X)

# Plot the result
plt.figure(figsize=(8, 6))
scatter = plt.scatter(X_dmt[:, 0], X_dmt[:, 1], c=y, cmap='viridis')

# Create legend
legend1 = plt.legend(*scatter.legend_elements(), title="Classes")
plt.gca().add_artist(legend1)  # Add the legend to the current axes

plt.title('DMT-EV visualization of Iris dataset')
plt.xlabel('DMT-EV Component 1')
plt.ylabel('DMT-EV Component 2')
plt.savefig('dmt.png')

You can alse separate the training and inference steps:

dmt.fit(X)
X_dmt = dmt.transform(X)

If you want to compare the results with other dimension reduction methods(t-SNE, UMAP), you can use the following code:

dmt.compare(X, "comparison.png")

Example Results

The code is in example.

Mnist Dataset

breast_cancer Dataset

20News Dataset

PBMC Dataset

Cite the paper

@article{zang2024dmt,
  title={DMT-EV: An Explainable Deep Network for Dimension Reduction},
  author={Zang, Zelin and Cheng, Shenghui and Xia, Hanchen and Li, Liangyu and Sun, Yaoting and Xu, Yongjie and Shang, Lei and Sun, Baigui and Li, Stan Z},
  journal={IEEE Transactions on Visualization and Computer Graphics},
  volume={30},
  number={3},
  pages={1710--1727},
  year={2024},
  publisher={IEEE}
}

License

DMT-EV is released under the MIT license.