MIMIC Benchmark reimplemented in PyTorch. For examples and documentation checkout: https://torchmimic.readthedocs.io/en/latest/
To install, run:
pip install torchmimic
This library contains PyTorch datasets that will load your existing MIMIC Benchmark data. To use those datasets, just import the data you want from torchmimic.data
:
from torchmimic.data import DecompensationDataset
from torchmimic.data import IHMDataset
from torchmimic.data import LOSDataset
from torchmimic.data import PhenotypingDataset
We've included some of the models used in the benchmark paper for reproduction and comparison with your own methods. Current models include:
- Standard LSTM
You can access these models from torchmimic.models
:
from torchmimic.models import StandardLSTM
Each of the benchmarks can be found in torchmimic.benchmarks
:
from torchmimic.benchmarks import DecompensationBenchmark
from torchmimic.benchmarks import IHMBenchmark
from torchmimic.benchmarks import LOSBenchmark
from torchmimic.benchmarks import PhenotypingBenchmark
At a minimum, the benchmark need a model to be trained. You can create the model from one of our existing ones or you can create your own.
This library has built-in logging of model configurations and key metrics using the Weights and Biases library.
- Phenotyping
- Length of Stay
- Decompensation
- In Hospital Mortality
- Multitask
- Standard LSTM
- Standard LSTM + Deep Supervision
- Logistic Regression
- Channelwise LSTM
- Channelwise LSTM + Deep Supervision
- Multitask LSTM
- Multitask Channelwise LSTM
- Test data for GitHub Workflow
If you use 'torchmimic' in your projects, please cite the following:
@software{torchmimic,
author = {Ryan King},
month = {10},
title = {{torchmimic}},
url = {https://github.com/github/linguist},
version = {0.0.3},
year = {2022}
}