Multimodal EHRMamba: Integrating Clinical Notes with State Space Models

Student: Divyansh Gangwar | Course: CS F434: Data Science for Healthcare Semester: First Semester 2025-2026 | Instructor: Prof. Manik Gupta

GitHub Repository & Reproducibility

This notebook serves as the main entry point for the project. The complete source code, including data pipelines and training scripts, is modularized in the src/ directory to ensure full reproducibility.

How to Run This Project

1. System Requirements:

• OS: Linux (Ubuntu 20.04+) or WSL2
• GPU: NVIDIA RTX 4060 (8GB VRAM) or better
• Python: 3.10 (Required for Mamba compilation)
• CUDA: 12.1

2. Installation:

STEP 1: Clone the repository

git clone https://github.com/aghori3004/EHRMamba_Multimodal.git
cd EHRMamba-Multimodal/

STEP 2: Create Environment (Strictly use Python 3.10)

python3.10 -m venv .venv
source .venv/bin/activate

STEP 3: Install Dependencies (PyTorch 2.1.2 + Mamba-SSM)

pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url [https://download.pytorch.org/whl/cu121](https://download.pytorch.org/whl/cu121)
pip install packaging ninja transformers pandas scikit-learn tqdm matplotlib seaborn

Note: Compiling Mamba takes

TORCH_CUDA_ARCH_LIST="8.9" pip install causal-conv1d>=1.2.0 mamba-ssm --no-cache-dir

3. Execution: We provide a master script to regenerate all artifacts (Data -> Models -> Metrics):

./run_pipeline.sh