import pandas as pd
import datetime as dt
import numpy as np

import plotly.express as px

url = 'https://raw.githubusercontent.com/neylsoncrepalde/projeto_eda_covid/master/covid_19_data.csv'

#transform date comlumns to datetime format together with the dataframe creation...
df = pd.read_csv(url, parse_dates = ['ObservationDate', 'Last Update'])

df.dtypes

df.head()

df.columns

import re

#re.sub(r'[/| ]', '') - will remove special characters(/ or empty spaces in this case).
def refine_cols(col_name):
    return re.sub(r'[/| ]', '', col_name).lower()

#changing column names to make thing easier.
df.columns = [refine_cols(i) for i in df.columns]
df.head()

df.countryregion.unique()

brasil = df.loc[(df.countryregion =='Brazil') & (df.confirmed > 0)]

px.line(brasil, 'observationdate', 'confirmed', title = 'Confirmed Cases in Brazil')

# Returns the daily covid cases
brasil['newcases'] = list(map(
    lambda row: 0 if (row==0) else brasil['confirmed'].iloc[row] - brasil['confirmed'].iloc[row - 1], np.arange(brasil.shape[0])
))

px.line(brasil, x = 'observationdate', y = 'newcases', title = 'Daily Covid Cases In Brazil')

import plotly.graph_objects as go
fig = go.Figure()

fig.add_trace(
    go.Scatter(x = brasil.observationdate, y= brasil.deaths, name = 'Mortes',
              mode = 'lines+markers', line = {'color': 'red'})
)

#layout
fig.update_layout(title = 'Deaths by COVID-19 in Brazil')

fig.show()

def growth_rate(data, var, start = None, end = None):
    
    #if start date is None set the first date avaible
    if start == None:
        start = data.observationdate.loc[data[var] > 0].min()
    else:
        start = pd.to_datetime(start)
    
    if end == None:
        end = data.observationdate.iloc[-1]
    else:
        end = pd.to_datetime(end)
    
    #past and present dates values
    past = data.loc[data.observationdate == start, var].values[0]
    present = data.loc[data.observationdate == end, var].values[0]
    
    #number of time periods we'll be analyzing
    n = (end - start).days
    
    #calculate rate
    rate = (present/past)**(1/n) - 1
    
    return rate*100

growth_rate(brasil, 'confirmed')

def daily_growth_rate(data, var, start=None):
    #if start date is None set the first date avaible
    if start == None:
        start = data.observationdate.loc[data[var] > 0].min()
    else:
        start = pd.to_datetime(start)
        
    end = data.observationdate.max()
    
    n = (end - start).days
    
    rates = list(map(
        lambda x: (data[var].iloc[x] - data[var].iloc[x-1]) / data[var].iloc[x - 1],
        range(1, n + 1)
    ))
    
    return np.array(rates) * 100

daily_rates = daily_growth_rate(brasil, 'confirmed')

first_day = brasil.observationdate.loc[brasil.confirmed > 0].min()

px.line(x = pd.date_range(first_day, brasil.observationdate.max())[1:],
        y = daily_rates, title = 'Growth Rates of Confirmed Cases in Brazil'
       )

from statsmodels.tsa.seasonal import seasonal_decompose
import matplotlib.pyplot as plt

confirmed_cases = brasil.confirmed
#setting observationdate as index
confirmed_cases.index = brasil.observationdate
confirmed_cases

res = seasonal_decompose(confirmed_cases)

rows = 2
cols = 2
fig = plt.figure(figsize =(20,5))
fig.tight_layout(pad = 1.0)

#current series
ax = fig.add_subplot(2,2,1)
ax.plot(res.observed)

#trends
ax = fig.add_subplot(2,2,2)
ax.plot(res.trend)

#seasonality
ax = fig.add_subplot(2,2,3)
ax.plot(res.seasonal)

#residual
ax = fig.add_subplot(2,2,4)
ax.plot(confirmed_cases.index, res.resid)
ax.axhline(0, linestyle = '--', c = 'black')
plt.show()    

from pmdarima.arima import auto_arima

model = auto_arima(confirmed_cases)

fig = go.Figure(go.Scatter(
                x = confirmed_cases.index, y = confirmed_cases, name = 'Observed'
               ))

fig.add_trace(go.Scatter(
    x = confirmed_cases.index, y=model.predict_in_sample(), name = 'Prediction'
))

fig.add_trace(go.Scatter(
    x = pd.date_range('2020-05-20', '2020-06-20'), y = model.predict(31), name = 'Forecast'
))

fig.update_layout(title= 'Confirmed cases forecast for the next 30 days in Brazil')
fig.show()

from prophet import Prophet

#preprocessing

#all data excluding the last 5 cases
train = confirmed_cases.reset_index()[:-5]


test = confirmed_cases.reset_index()[-5:]

#Renaming Columns
train.rename(columns = {'observationdate':'ds', 'confirmed' : 'y'}, inplace = True)
test.rename(columns = {'observationdate':'ds', 'confirmed': 'y'}, inplace = True)

#Defining Growth Model
prophet = Prophet(growth = 'logistic', changepoints = ['2020-03-21','2020-03-30','2020-04-25', '2020-05-03','2020-05-10'])


pop = 211463256
train['cap'] = pop

#training
prophet.fit(train)

#predictions
future_dates = prophet.make_future_dataframe(periods =200)
future_dates['cap']=pop
forecast = prophet.predict(future_dates)

fig = go.Figure()

fig.add_trace(go.Scatter(x=forecast.ds, y=forecast.yhat, name='Prediction'))
fig.add_trace(go.Scatter(x=train.ds, y=train.y, name='Observer - Train'))
fig.update_layout(title = 'Prediction on confirmed cases in Brazil')
fig.show()