Optimization of DeepGLEAM on Flu Forecasting Time-Series Data

The current COVID-19 pandemic and common flu highlight the importance of time-sensitive information in biomedical institutions, politics, and economics. The application of data science in creating real-time predictive models is crucial to help researchers and world leaders better understand disease spread and take preventative measures.

GLEAM Prediction before and after Interpolation

ARIMA and ETS

Prediction

Four weeks ahead Flu prediction residual between groundtruth and prediction for 10 states

Result Comparison

Setup, Model training and Model Testing

1. Requirements

>>> pip install -r requirements.txt

2. Train models and make prediction (Model already trainned in the submission)

>>> python3 run.py --config_filename=data/model/dcrnn_cov.yaml

3. For Test, run the following command

>>> ./test.sh

• Visualization

  • After running the command for test, a new folder named plot_weeknumber_result will appear containing [0.025, 0.5, 0.975] residual predictions the .npz files
  • Select the one with lowest MAE score
  • Run the flu_forecast_result_plot notebook

Docker

>>> docker build -f ./Dockerfile -t Dockerfile .
>>> docker run --rm -it Dockerfile /bin/bash
>>> launch.sh -i xiangyikong/dsc180a:latest #Use this command below to launch the image in DSMLP