Ventilator Pressure Prediction: A Deep Learning Approach This project provides a complete, research-level pipeline for the Kaggle "Ventilator Pressure Prediction" competition. It includes advanced feature engineering, training of both Transformer and LSTM models, hyperparameter tuning, and a final weighted ensemble with post-processing to generate a submission.

Project Structure The project code is organized in the src/ directory. The data and saved models are expected to be in a separate, untracked directory as shown below.

Setup and Installation

1. Prerequisites Python 3.8+

An NVIDIA GPU with CUDA is highly recommended for training.

2. Environment Setup

Create and activate a virtual environment

python3 -m venv venv source venv/bin/activate

Install the required libraries

pip install -r requirements.txt

3. Kaggle API Setup & Data Download This project requires the Kaggle API to download the dataset.

Get your API credentials from your Kaggle account page and place the kaggle.json file in ~/.kaggle/.

Create the data directory:

mkdir -p ventilator-pressure-prediction/data

Download and unzip the data:

From the root of the project directory

kaggle competitions download -c ventilator-pressure-prediction -p ventilator-pressure-prediction/data/ unzip ventilator-pressure-prediction/data/ventilator-pressure-prediction.zip -d ventilator-pressure-prediction/data/

This will place train.csv, test.csv, and sample_submission.csv inside the correct data/ folder.

Running the Pipeline The entire project can be run from the main.py script. The script uses absolute paths defined in src/config.py, so ensure they are correct for your system.

From the root of the project directory

python main.py