This repository contains several models for a regression of the Google Brain - Ventilator Pressure Prediction (VPP) dataset. The website for this Kaggle Competition can be found here.
Neural networks are used for feature extraction and regression. These were implemented in Python using the TensorFlow library.
Furthermore, not only are various neural networks available, but also training procedures and scripts for data preprocessing and storage.
The aim of this competition is to predict the airway pressure in the breathing circuit of a breath based on numerous time series of breaths. For this purpose, various control inputs and lung property are available, which serve as a basis for prediction. The competition website also lists further details on the data.
The data cannot be made available here. However, they can be downloaded from the Kaggle website (link above).
The following models were trained and tested:
- Small Gated Recurrent Units (GRU), this is a Recurrent Neural Network (RNN) model
- Big GRU model
- Big Long Short-Term Memory (LSTM) model
| Files | Description |
|---|---|
| Dataset/ | contains the data and the submissions |
| Models/ | contains the trained models |
| Plots/ | contains all plots from the training and testing |
| dataset.py | provides the dataset and prepares the data |
| helpers.py | provides auxiliary classes and functions for neural networks |
| Job.sh | provides a script to carry out the training on a computer cluster |
| models.py | provides the models |
| train.py | provides functions for training and testing |
| vpp_jupyter.ipynb | contains all existing python files in a jupyter notebook |
The scores were calculated by Kaggle. The metric is the mean absolute error (MAE).
| Models | Private leaderboard score | Training time (hh:mm:ss) |
|---|---|---|
| GRU | 0.3943 | 02:27:43 |
| Big GRU | 0.2917 | 07:38:54 |
| Big LSTM | 0.3021 | 10:57:58 |