Saifahmadkhan / PlugNPlay

This library is a PlugNPlay version of our novel pipeline VacSIM. We have built a pipeline that makes Reinforcement Learning guided policies for optimal distribution of Covid vaccines.

VacSIM-PlugNPlay

This library is a PlugNPlay version of our novel pipeline VacSIM.

Overview

A COVID-19 vaccine is our best bet for mitigating the ongoing onslaught of the pandemic. However, vaccine is also expected to be a limited resource. An optimal allocation strategy, especially in countries with access inequities and temporal separation of hot-spots, might be an effective way of halting the disease spread. We approach this problem by proposing a novel pipeline VacSIM which follows reinforcement learning based approach for optimal distribution of vaccines for Covid-19.

Quickstart

Clone the github repo.

!git clone https://github.com/Saifahmadkhan/PlugNPlay

Select tenserflow version 1.x

%tensorflow_version 1.x

Install the requirements.

!pip install -r PlugNPlay/requirements1.txt
!sed 's/#//' PlugNPlay/requirements2.txt | xargs apt install

Import DataLoader and Pipeline classes.

from PlugNPlay.DataLoader import DataLoader
from PlugNPlay.Pipeline import Pipeline

Load context data into a pandas dataframe.

import pandas as pd
df=pd.read_csv('projections.csv')
df.head()

Date	State / UT	Confirmed cases	Death Rate	Recovery Rate	Population	Susceptible	Overall Hospital Beds	Overall ICU Beds	Ventilators	Age Distribution
2020-12-01	Assam	2.188312e+05	0.011089	0.954143	31205576	3.098674e+07	24178	1209	604	4127732
2020-12-01	Delhi	5.574192e+05	0.011039	0.905071	16787941	1.623052e+07	39455	1973	986	2343806
2020-12-01	Jharkhand	1.111680e+05	0.011081	0.948757	32988134	3.287697e+07	26496	1325	662	4528093
2020-12-01	Maharashtra	1.860920e+06	0.010928	0.906055	112374333	1.105134e+08	231739	11587	5793	19912066
2020-12-01	Nagaland	1.210262e+04	0.010670	0.812338	1978502	1.966399e+06	2561	128	64	211983

Create DataLoader object by passing context dataframe, column number corresponding to date, column number corresponding to regions/candidates, column number corresponding to susceptible population and column numbers where context values are not normalized(sum over all the regions/candidates is not equal to 1).

data_obj=DataLoader(df,date_column=0,candidates_column=1,susceptible_column=6,unnormalized_cols=[2,6,5,7,8,9,10])

Create Pipeline object by passing DataLoader object and model type. Model types allowed are 'DQN' and 'ACKTR'.

pipeline_obj=Pipeline(data_object=data_obj,model_type='DQN')

Run the pipeline and get the distributions in output.csv.

pipeline_obj.run()

Citation

If you use VacSIM for your research, please consider citing the following paper:

VacSIM: Learning Effective Strategies for COVID-19 Vaccine Distribution using Reinforcement Learning.

Raghav Awasthi, Keerat Kaur Guliani, Saif Ahmad Khan, Aniket Vashishtha, Mehrab Singh Gill, Arshita Bhatt, Aditya Nagori, Aniket Gupta, Ponnurangam Kumaraguru, Tavpritesh Sethi.