dmt-learn 0.0.81

dmt learn package

DMT-HI: MOE-based Hyperbolic Interpretable Deep Manifold Transformation for Unspervised Dimensional Reduction

(Our Paper)[https://arxiv.org/abs/2410.19504]

The code includes the following modules:

• Datasets (Mnist, CIFAR-10, HCL, 20newsgroups)
• Training for DMT-HI
• Evaluation metrics
• Visualisation
• Explainable Analyses

Configurating python environment

We recommend using conda for configuration. You can refer to our install-env.sh to configure the environment.

conda create -n nml python=3.9
conda activate nml
bash install_env.sh

Dataset

This project utilizes several datasets, including 20NG, HCL, MNIST, and CIFAR-10. Please follow the instructions below to understand the dataset structure and usage.

1. 20NG Dataset

The 20NG dataset is already included in this GitHub repository.

2. HCL Dataset

The HCL dataset must be manually downloaded from the following link: Download HCL Dataset. Once downloaded, please place the file HCL60kafter-elis-all.h5ad into the data_path/ directory.

3. MNIST and CIFAR-10 Datasets

The MNIST and CIFAR-10 datasets do not require manual download. These datasets will be automatically downloaded upon the first execution of the project. Please ensure that you have a stable internet connection during the first run to automatically download these datasets.

Run DMT-HI

You can run DMT-HI with a single line of code to get latent embedding.

Minimun replication

Running minimal replication can be done with the following command:

python main.py fit -c=conf_new/nml4/mnist.yaml

Analyses

After successfully running DMT-HI for the first time, you can use the built-in analyzer to further explore the results. Follow the steps below to configure and run the analyzer.

Steps to Use the Analyzer

1. Open dash_main.py File:
   Navigate to the file dash_main.py in the project directory.

2. Modify the Model and Image Paths:
   In dash_main.py, update the following lines to match the model and image outputs generated by DMT-HI:

   • Line 14: Set the path to the saved model generated by DMT-HI.
   • Line 17: Set the path to the saved images generated during the model run.

3. Run the Analyzer:
   After making the necessary changes, run the following command to start the analyzer:

   python dash_main.py
   

4. Access the Analyzer:
   Once the script is running, it will return a local URL. Open the URL in your web browser to access the DMT-HI Analyzer.

Through this web-based interface, you can visualize latent embeddings.