COVID19 Travel (Un)planner

Hackathon project submission repository for Pod 0.3.1 (Lazy Lobsters) Team 1

COVID19 Travel (Un)planner

Project Description

This project aims to use SciML to fit a differential equation to the COVID-19 Number-Of-Cases time series data. This would allow us to predict the number of cases in the future, to some degree of accuracy. This would be done separately per district or state, as per the availability of reliable data. The results would be displayed on an interactive graphical webpage.

The project also implements an interesting use case of the prediction model: a mobile application that is a personal scheduler/calendar. The user can enter where they are headed to in the next few days, and the application would warn them if those regions are predicted to have a high number of cases then.

Data sourced from https://api.covid19india.org/.

Source organization

Server code is present in Server directory
Datasets and the preprocessing script used are present in Dataset directory
Trained models are in BSON format in the Server/data directory
Source of react native application is present in the Application/ directory with the build files (APK et al) in Application/dist directory

Web Application Homepage

Tech Stack

Julia : For the backend
React Native : For the Application
DiffEqFlux : For training district wise models

Mobile Application

React Native
Expo

Local Setup

Server

Clone repository

	git clone https://github.com/MLH-Fellowship/0.3.1-team1-COVIDTravelPlanner && cd 0.3.1-team1-COVIDTravelPlanner

Install the server dependencies

	julia --project=./Server -e "using Pkg; Pkg.instantiate()"

Start the server

	julia --project=./Server Server/server.jl

Mobile App using Expo

Change into the mobile-app repository

cd mobile-app

Install dependencies

npm install

Run the Expo development server

npm start

Follow the instructions on screen

Predictive Models for Covid19

We provide the code for training the neural network models and running inference using them in the Covid19Modelling module inside Server/lib. First we need to install the dependencies using

	julia --project=./Server -e "using Pkg; Pkg.instantiate()"

Training the models

Our models need to be retrained daily / whenever a new data point is available. To train the model on a single district run the following code:

include("Server/lib/Covid19Modelling.jl") # Import the module

# The dataset needs to be updated on a daily basis
Covid19Modelling.train_named_locations("Dataset/final.csv", [<Names of Districts>], [<Names of States>], retrain = true, each_epochs = 100)

If all the models need to be retrained, simply change the last line above to the following

Covid19Modelling.train_alll_districts("Dataset/final.csv", retrain = true, each_epochs = 100)

Getting the model predictions

Once the models are trained to obtain the predictions for the next 7 days, we do the following:

# Gives the predictions for the next 7 days
Covid19Modelling.prediction(<Name of state>, <Name of District>, [1, 2, 3, 4, 5, 6, 7])

# Gives the predictions for the 2nd, 4th and 6th day from today
Covid19Modelling.prediction(<Name of state>, <Name of District>, [2, 4, 6])

# Gives the predictions for day after tomorrow
Covid19Modelling.prediction(<Name of state>, <Name of District>, 2)

Basically, you can get the predictions for any day in the next week. The function returns the total active cases and the total resolved cases (recovered + deceased)

Model Description

We model our Differential Equations as follows using the DifferentialEquations.jl packages

We draw inspiration from the previous works [1] and [2]. We use a multilayer perceptron to parameterize the differential equations and using backpropogation through the solver we are able to learn the parameters in an end-to-end manner. To learn to predict the case counts we use minimize the weighted squared difference between the log of the case numbers.

References

[1] Rackauckas, Christopher, et al. "Universal Differential Equations for Scientific Machine Learning." arXiv preprint arXiv:2001.04385 (2020).

[2] Dandekar, Raj, and George Barbastathis. "Quantifying the effect of quarantine control in Covid-19 infectious spread using machine learning." medRxiv (2020).

ETS-ReactNative1/0.3.1-team1-COVIDTravelPlanner

COVID19 Travel (Un)planner