covid.Rmd

---
title: "Covid-19"
output: 
  flexdashboard::flex_dashboard:
    orientation: rows
    vertical_layout: scroll
    theme: lumen
---


```{r setup, include=FALSE}
library(dplyr)
library(prophet)
library(rBayesianOptimization)
library(ggplot2)
library(flexdashboard)
#options(error=recover)

azn = data.table::fread('https://covid.ourworldindata.org/data/owid-covid-data.csv') %>% 
  mutate(date = as.Date(date)) %>% filter(location %in% c('Azerbaijan'))
index_max = nrow(azn)

df= azn %>% select(date,new_cases) %>% rename(ds=date,y=new_cases)
#df <- zoo::na.locf(df)
#df$y <- df$y + 1
#df$y[df$y==0]=NA

max_date = max(df$ds)
min_date = max_date - 4

df_ = df %>% filter(ds < min_date)

prophet_fit = function(n.changepoints, changepoint.range ,   seasonality.prior.scale, 
                       holidays.prior.scale , changepoint.prior.scale , 
                       mcmc.samples , interval.width , uncertainty.samples) {
  
  
  m = prophet(df_,growth = "linear", 
              yearly.seasonality = TRUE, 
              weekly.seasonality = TRUE, daily.seasonality = TRUE, 
              seasonality.mode = "additive",
              n.changepoints = n.changepoints, changepoint.range = changepoint.range, seasonality.prior.scale = seasonality.prior.scale, 
              holidays.prior.scale = holidays.prior.scale, changepoint.prior.scale = changepoint.prior.scale, 
              mcmc.samples = mcmc.samples, interval.width = interval.width, 
              uncertainty.samples = uncertainty.samples)
  
  future <- make_future_dataframe(m, periods = 20)

  forecast <- predict(m, future) %>% mutate(yhat=ifelse(yhat<0,
                                                        50, 
                                                        yhat))
  
  actual = df %>% filter(ds > min_date) %>% pull(y)
  yhat = forecast %>% filter(ds > min_date, ds <= max_date) %>% pull(yhat)
  
  
  score_ <- list(Score = -Metrics::mae(actual,yhat),
                 Pred = 0)
  
  score_
  
}

search_bound_prophet <- list(n.changepoints= c(0L, 150L), 
                             changepoint.range= c(0.6, 0.9) , 
                             seasonality.prior.scale= c(2, 20), 
                             holidays.prior.scale= c(2, 20) , changepoint.prior.scale= c(2, 20) , 
                             mcmc.samples=c(10,50) , interval.width=c(0,1) , uncertainty.samples=c(1000,1e4)
)

search_grid_prophet <- data.frame(
                                  n.changepoints= runif(5,0L, 150L), 
                                  changepoint.range= runif(5, 0.6, 0.9) , 
                                  seasonality.prior.scale= runif(5,2,20), 
                                  holidays.prior.scale= runif(5,2,20) , changepoint.prior.scale= runif(5,2,20) , 
                                  mcmc.samples=runif(5,10,50) , interval.width=runif(5,0,1) , 
                                  uncertainty.samples=runif(5,1000,1e4)
)
#head(search_grid_prophet)


bayes_prophet <- BayesianOptimization(FUN = prophet_fit, bounds = search_bound_prophet,
                                     init_points = 2, init_grid_dt = search_grid_prophet,
                                     n_iter = 12, acq = "ucb")


print(bayes_prophet$Best_Par)

obj=bayes_prophet$Best_Par

m = prophet(df,growth = "linear", 
            yearly.seasonality = TRUE, 
            weekly.seasonality = TRUE, daily.seasonality = TRUE, 
            seasonality.mode = "additive",
            n.changepoints = obj[1], changepoint.range = obj[2], seasonality.prior.scale = obj[3], 
            holidays.prior.scale = obj[4], changepoint.prior.scale = obj[5], 
            mcmc.samples = obj[6], interval.width = obj[7], 
            uncertainty.samples = obj[8])

future <- make_future_dataframe(m, periods = 60)
#tail(future)
forecast <- predict(m, future) %>% mutate(yhat=ifelse(yhat<0,
                                                      1, 
                                                      yhat))
#tail(forecast[c('ds', 'yhat', 'yhat_lower', 'yhat_upper')])


actual = df %>% filter(ds > min_date) %>% pull(y)
yhat = forecast %>% mutate(ds=as.Date(ds)) %>%  filter(ds > min_date, ds <= max_date) %>% pull(yhat)

Metrics::mae(actual,yhat)



data = forecast %>% select(ds,yhat) %>% left_join(df) %>% mutate(ds=as.Date(ds),
                                                                 yhat=as.integer(yhat)) #%>%filter(ds>='2020-08-01')
names(data) = c('Date','Prediction','Actual')


p=ggplot() + 
  geom_line(data = data, aes(x = Date, y = Actual), colour = "darkgreen") +
  geom_line(data = data, aes(x = Date, y = Prediction), colour = "red") +
  scale_x_date(date_breaks = "1 month",# date_minor_breaks = "1 week",
               date_labels = "%B %Y") +
  xlab('Dates') +
  ylab('Daily cases')+ggtitle("Green = Actual \nRed = Predicted")

```



Azerbaijan
=====================================  

Row {data-height=130}
-----------------------------------------------------------------------

### Date

```{r}
calen <- azn$date[index_max]
valueBox(calen, icon = "fa-calendar-alt",color = "danger")
```

### Total Cases

```{r}
tot_cases <- azn$total_cases[index_max]
valueBox(prettyNum(tot_cases, big.mark = ','), icon = "fa-users")
```

### Total Deaths

```{r}
tot_deaths <- azn$total_deaths[index_max]
valueBox(prettyNum(tot_deaths, big.mark = ','), icon = "fa-exclamation-triangle")
```

### New cases

```{r}
new_cases <- azn$new_cases[index_max]
valueBox(prettyNum(new_cases, big.mark = ','), 
         icon = "fa-diagnoses",
         color = "warning")
```

### New Deaths

```{r}
new_deaths <- azn$new_deaths[index_max]
valueBox(prettyNum(new_deaths, big.mark = ','), 
         icon = "fa-exclamation-circle",
         color = "danger")
```


Row {data-height=400}
-------------------------------------

### Prediction for the next 2 months

```{r}
plotly::ggplotly(p)
```