Covid-19, Analysis, Visualization, Prediction

Author: Zhuofei, Zhou

---

Introduction

Covid-19

Novel Coronavirus 2019, on 12 January 2020, WHO officially named it 2019-NCOV. Coronaviruses are a large family of viruses known to cause colds and more serious illnesses such as Middle East Respiratory syndrome (MERS) and severe acute respiratory syndrome (SARS). Novel Coronavirus is a novel coronavirus strain that has never been found in humans before. More information is available.百度百科

data

Data from COVID-19 Data Repository by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University.link

• time series covid19 confirmed global.csv.(github link).

• time series covid19 deaths global.csv.(github link)

• time serise covid19 recoverd global.csv(github link)

• cases contry.csv.(github link)

• cases time.csv. (github link)

download here:

---

import some packages in julia:

---

Analysis

Analysis Global Data

Now, we just use the data to rebuild a new dataframe to store the Global data. I create a new dataframe named as Global_num to store the number of deaths and comfirmed numbers，time from 2020-01-22 to 2020-10-02.

Data Operate

operating as folows:

Plot show

Then i plot the number of comfirmed cases and deaths cases over time.

As we can see, the number of comfirmed cases continues to increase exponentially. But the number of deaths cases seems to increase linearly.

Now, start calculate the New cases every day，and the deaths rate.

the maximum death rate in 2020-4-29

So, the new comfirmed cases continues to rise, the death rate is already falling.

Prediction

Global Trend:

It is useful to understand the global trend of an increase in the number of cases over time. There is always a pattern in any data, but the concern is how strongly data follows a pattern. COVID-19 spreads exponentially.

i think of some ways to estimate the curve:

• Numerical Analysis: ❎

• linear regression, after transmission: ✅

• neural network: ✅

So, we first focus on comfirmed case:

do $y^{\prime }=\sqrt{y}$ when i use to transition. maybe will get a linea function.:

Other ways: Neural Network

import Flux to do meachine learning.

the all code here:

using Flux, Statistics
using Flux.Data: DataLoader
using Flux: throttle
using Parameters: @with_kw
using DelimitedFiles
using IterTools: ncycle
using Dates
using DataFrames
using CSV
using Plots


@with_kw mutable struct Args
    η::Float64 = 0.001
    batchsize::Int = 1
    epochs::Int = 1000
end
cd(@__DIR__)
pwd()
#read data:
dt, Header = readdlm("time_series_covid19_comfirmed_global.csv", ',', header=true)
dt_deaths, Header_deaths = readdlm("time_series_covid19_deaths_global.csv", ',', header=true)
dt_recover,Header_recov = readdlm("time_series_covid19_recovered_global.csv",',',header=true)


#create a new datafram to store datum
comfirmed_num = []
deaths_num = []
recovered_num = []
for i in 5:259
    push!(comfirmed_num, sum(dt[:, i]))
    push!(deaths_num, sum(dt_deaths[:, i]))
    push!(recovered_num, sum(dt_recover[:,i]))
end
dates = Date(2020, 1, 22):Day(1):Date(2020, 10, 2)
Global_num = DataFrame(Date=dates, 
    comfirmed_num=comfirmed_num, 
    deaths_num=deaths_num, 
    recovered_num=recovered_num)

#read train_data
n, = size(comfirmed_num)
x = 0:n-1
y = Global_num.comfirmed_num
args = Args()
train_data = DataLoader((Array(x), Float64.(y)), batchsize=args.batchsize)

# define leaky relu
Lelu(x, α=100) = (x ≥ 0 ? x : x/α)

#define Model
m = Chain(
    Dense(1, 40, Lelu),
    Dense(40, 40, Lelu),
    Dense(40, 40, Lelu),
    Dense(40, 1, Lelu)
)

#define loss function
loss(x, y) = Flux.mse(m(x), y)

#define parameters
ps = Flux.params(m)

#define Opt
opt = ADAM(args.η)

#train model
Flux.train!(loss, ps, ncycle(train_data, args.epochs), opt)

#visualize
flux_y = []
for i in Array(x)
    push!(flux_y, Array(m([i]))[1])
end
flux_y = Float64.(flux_y)
plot(dates, [flux_y Float64.(y)], 
    label=["predict" "real"], xlabel="date", ylabel="Numbers", size=(900, 600))