def object_minimize(x,t,cumdata_cases,cum_deaths):
SUCQ = odeint(sucq, [x[9],x[10],x[11],x[12],x[13],x[14],x[15],0,x[16]],t, args=(x[0],x[1],x[2],x[3],x[4],x[5],x[6],x[7],x[8]))
w1=1
w2=1
r1 = -1*hy.nse(SUCQ[:,3],cumdata_cases)
r2 = -1*hy.nse(SUCQ[:,7],cum_deaths)
return (w1*r1+w2*r2)/(w1+w2)
def metrics_computation(data_list):
"""Calculates the metrics values based on the model output. (RMSE,
MSE, MAE and R²)
Parameters
----------
data_list : list with the separated data corresponding to each
station.
Returns
-------
m_list : list with the separated data corresponding to each
station with metrics included.
"""
m_list = []
for el in data_list:
el["rmse"], el["mse"], el["mae"], el["r2"] = 0, 0, 0, 0
el["vaf"], el["nse"] = 0, 0
lines = el.shape[0]
for i in range(lines):
yt = np.asarray(el.iloc[i, 6], dtype=np.float32)
yp = np.asarray(el.iloc[i, 7], dtype=np.float32)
el.iloc[i, 10] = metrics.mean_squared_error(yt, yp, squared=False)
el.iloc[i, 11] = metrics.mean_squared_error(yt, yp, squared=True)
el.iloc[i, 12] = metrics.mean_absolute_error(yt, yp)
el.iloc[i, 13] = metrics.r2_score(yt, yp)
el.iloc[i, 14] = vaf_metric(yt, yp)
el.iloc[i, 15] = he.nse(yt, yp)
m_list.append(el)
return m_list
def object_minimize(x,t,cumdata_cases):
SUCQ = odeint(sucq, [x[3],x[4],x[5],x[6]],t, args=(x[0],x[1],x[2]))
#
# if cumdata_cases[-1:] < 1000:
# return sum( ( np.log10(cumdata_cases)-np.log10(SUCQ[:,3].ravel()) )**2)
# if cumdata_cases[-1:] >= 1000:
# return sum((cumdata_cases-SUCQ[:,3].ravel())**2)
return -1*hy.nse(SUCQ[:,3],cumdata_cases)
def Ajust_SUCQ(FILE,pop,extrapolação,day_0,variavel,pasta):
data_covid = pd.read_csv(pasta+"/"+FILE, header = 0, sep = ";")
data_covid = data_covid[['DateRep',variavel]]
nome_data = 'DateRep'
df = data_covid
day_0_str=df[nome_data][0][:4]+'-'+df[nome_data][0][5:7]+'-'+df[nome_data][0][-2:]
#day_0_str = data_covid['DateRep'][0][-4:]+'-'+data_covid['DateRep'][0][3:5]+'-'+data_covid['DateRep'][0][:2]
date = np.array(day_0_str, dtype=np.datetime64)+ np.arange(len(data_covid))
if date[0]>=np.array(day_0, dtype=np.datetime64):
t0 = 0
else:
dif_dias =np.array('2020-03-18', dtype=np.datetime64)-date[0]
t0 = dif_dias.astype(int)
date= date[t0:]
cumdata_covid = data_covid[['Cases']].cumsum()
cumdata_cases = cumdata_covid['Cases'].values[t0:]
days_mens = np.linspace(1,len(cumdata_cases),len(cumdata_cases))
p0,bsup,binf = bounds(FILE)
p0.append(cumdata_cases[0])
bsup.append(cumdata_cases[0]+10**-9)
binf.append(cumdata_cases[0]-10**-9)
popt = min_minimize(cumdata_cases,C,p0,days_mens,bsup,binf)
r,p,K,Co = popt
solution = C(days_mens, r, p, K,Co)
NSE = hy.nse(solution,cumdata_cases)
RMSE = hy.rmse(solution,cumdata_cases)
MARE = hy.mare(solution,cumdata_cases)
print(FILE[9:-4])
print("r = %f " % (r))
print("p = %f " % (p))
print("K = %f " % (K))
print("NSE = %.5f " % (NSE))
print("#######################")
days_mens = np.linspace(1,len(cumdata_cases)+365,len(cumdata_cases)+365)
solution = C(days_mens, r, p, K,Co)
date_future = np.array(date[0], dtype=np.datetime64)+ np.arange(len(date)+extrapolação)
saída = pd.DataFrame(solution, columns=["Cases"])
saída["date"] = date_future
path_out = "C:/Users/ravel/OneDrive/Área de Trabalho/DataScientist/sklearn/COVID-19/CasosPorEstado/GLM_covid19/data/data_simulated"
saída.to_csv(path_out+"/"+FILE,sep=";")
return [r,p,K,NSE,RMSE,MARE]
def Ajust_SUCQ(data_covid, pop, passo, j):
data_covid = data_covid[['DateRep', 'Cases']]
t0 = j
tf = j + passo
cumdata_covid = data_covid[['Cases']].cumsum()
cumdata_cases = cumdata_covid['Cases'].values[t0:tf]
# data=data_covid[['DateRep']].values[t0:tf]
# day_last = data[-1:][0][0]
data = data_covid[['DateRep']].values[t0:tf][-1:][0][0]
day_last_str = data[:4] + '-' + data[5:7] + '-' + data[-2:]
day_last = day_last_str #np.array(day_last_str, dtype=np.datetime64)
N = pop * 10**6
t = np.linspace(1, len(cumdata_cases), len(cumdata_cases))
So, Uo, Qo, Co = [
.9 * N, 6 * cumdata_cases[0], cumdata_cases[0], cumdata_cases[0]
] # padrão [.8*N,6*cumdata_cases[0],cumdata_cases[0],cumdata_cases[0]]
alfa_0, beta_0, gama1_0 = [.2 / So, .3, .1] # padrão [.5/N,.1,.19]
p0 = [alfa_0, beta_0, gama1_0, So, Uo, Qo, Co]
bsup = [0.9 / So, .50, .20, N, Uo * 2., Qo * 2.0, Co + 10**-9]
binf = [0.1 / So, .05, .01, .7 * N, Uo * .5, Qo * 0.5, Co - 10**-9]
#p0 = ajust_curvefit(days_mens,cumdata_cases,p0,bsup,binf)
popt = min_minimize(cumdata_cases, sucq_solve, p0, t, bsup, binf)
alfa_0, beta_0, gama1_0, So, Uo, Qo, Co = popt
solution = SUCQ(t, alfa_0, beta_0, gama1_0, So, Uo, Qo, Co)
NSE = hy.nse(np.log10(cumdata_cases), np.log10(solution[:, 3]))
if NSE >= 0.25:
return [alfa_0 * So / gama1_0, day_last]
else:
return [np.nan, day_last]
def Ajust_SUCQ(FILE,pop,extrapolação):
path = 'C:/Users/ravellys/Documents/GitHub/COVID-19-Brasil/COVID-19-Brasil/data/DADOS/'
data_covid = pd.read_csv(path+FILE, header = 0, sep = ";")
data_covid=data_covid[['DateRep','Cases']]
nome_data = 'DateRep'
df = data_covid[['DateRep','Cases']]
day_0_str=df[nome_data][0][:4]+'-'+df[nome_data][0][5:7]+'-'+df[nome_data][0][-2:]
date = np.array(day_0_str, dtype=np.datetime64)+ np.arange(len(data_covid))
if date[0]>=np.array('2020-05-09', dtype=np.datetime64):
t0 = 0
else:
dif_dias =np.array('2020-05-09', dtype=np.datetime64)-date[0]
t0 = dif_dias.astype(int)
date= date[t0:]
cumdata_covid = df[['Cases']].cumsum()
cumdata_cases = cumdata_covid['Cases'].values[t0:]
days_mens = np.linspace(1,len(cumdata_cases),len(cumdata_cases))
N = pop*10**6
t = days_mens
So,Uo,Qo,Co = [.9*N,6*cumdata_cases[0],cumdata_cases[0],cumdata_cases[0]] # padrão [.8*N,6*cumdata_cases[0],cumdata_cases[0],cumdata_cases[0]]
alfa_0,beta_0,gama1_0= [.2/So,.3,.1] # padrão [.5/N,.1,.19]
p0 = [alfa_0,beta_0,gama1_0,So,Uo,Qo,Co]
bsup = [0.4/So,.50,.20, N,Uo*2.,Qo*2.0,Co+10**-9]
binf = [0.09/So,.05,.01,.7*N,Uo*.5,Qo*0.5,Co-10**-9]
#p0 = ajust_curvefit(days_mens,cumdata_cases,p0,bsup,binf)
popt = min_minimize(cumdata_cases,sucq_solve,p0,t,bsup,binf)
alfa_0,beta_0,gama1_0,So,Uo,Qo,Co = popt
solution = SUCQ(t,alfa_0,beta_0,gama1_0,So,Uo,Qo,Co)
NSE = hy.nse(solution[:,3],cumdata_cases)
RMSE = hy.rmse(solution[:,3],cumdata_cases)
MARE = hy.mare(solution[:,3],cumdata_cases)
print(FILE[9:-4])
print("alfa = %f " % (alfa_0*So))
print("beta = %f " % (beta_0))
print("gamma = %f " % (gama1_0))
print("R = %f" %(alfa_0*So/gama1_0))
print("NSE = %.5f " % (NSE))
print("#######################")
date_future = np.array(date[0], dtype=np.datetime64)+ np.arange(len(date)+extrapolação)
days_future = np.linspace(1,len(cumdata_cases)+extrapolação,len(cumdata_cases)+extrapolação)
Cum_cases_estimated = SUCQ(days_future, *popt)
estimativafutura_saída=pd.DataFrame(Cum_cases_estimated[:,3], columns = ["Cases"])
estimativafutura_saída["S"]=Cum_cases_estimated[:,0]
estimativafutura_saída["U"]=Cum_cases_estimated[:,1]
estimativafutura_saída["Q"]=Cum_cases_estimated[:,2]
estimativafutura_saída["I"] = Cum_cases_estimated[:,1]+Cum_cases_estimated[:,2]+Cum_cases_estimated[:,3]
estimativafutura_saída["date"] = date_future
fileout = 'C:/Users/ravellys/Documents/GitHub/COVID-19-Brasil/COVID-19-Brasil/data/DADOS_estimados/'
estimativafutura_saída.to_csv(fileout+FILE,sep=";")
#par_SUQC_head = ["estado","população","data inicial","data final", "R","alfa","beta","gamma","So","Uo","Qo","Co","NSE"]
#par_SUQC = {"estado":FILE[9:-4],"população":pop,"data inicial":date[0],"data final": date[-1:][0],"R": alfa_0*N/gama1_0,"alfa":popt[0]*N,"beta":popt[1],"gamma":popt[2],"So":popt[3],"Uo":popt[4],"Qo":popt[5],"Co":popt[6],"NSE":NSE}
#append_dict_as_row("C:/Users/ravel/OneDrive/Área de Trabalho/DataScientist/sklearn/COVID-19/par_SUCQ.csv",par_SUQC,par_SUQC_head)
return [alfa_0*So/gama1_0,alfa_0*So, beta_0, gama1_0, So,Uo,Qo,Co, NSE, RMSE, MARE]
def Sensitivity(mypath, FILE, pop):
print(FILE)
t0 = 0
data_mensured = pd.read_csv(mypath + "/" + FILE, header=0, sep=";")
data_mensured = data_mensured[['DateRep', 'Cases']]
cumdata_covid = data_mensured[['Cases']].cumsum()
cumdata_cases = cumdata_covid['Cases'].values[t0:]
t = np.linspace(1, len(cumdata_cases), len(cumdata_cases))
for i in população:
if i[0] == FILE[9:-4]:
pop = float(i[1])
problem = {
'num_vars':
4,
'names': ['r', 'p', 'K', 'alfa'],
'bounds': [[0.001, 50], [0., 1.0],
[np.log10(0.01 * pop * 10**6),
np.log10(.5 * pop * 10**6)], [-10, 1]]
}
nsamples = 5000
param_values = saltelli.sample(problem, nsamples)
Y = []
W = []
for i in param_values:
r, p, K, alfa = i
alfa = 10**alfa
K = 10**K
Co = cumdata_cases[0]
cases_simulated = C(t, r, p, K, 10**alfa, Co)
NSE = hy.nse(cases_simulated, cumdata_cases)
if NSE <= 1:
NSE = NSE
else:
NSE = -1000000000
if NSE >= 0.75:
W.append([r, p, K, alfa, NSE])
Y.append([r, p, K, alfa, NSE])
Y = np.array(Y)
W = np.array(W)
print(W)
df_Y = pd.DataFrame(Y, columns=['r', 'p', 'K', 'alfa', 'NSE'])
df_W = pd.DataFrame(W, columns=['r', 'p', 'K', 'alfa', 'NSE'])
Si = sobol.analyze(problem, df_Y["NSE"].values)
df_S = pd.DataFrame(Si["S1"], columns=["S1"])
df_S["ST"] = Si["ST"]
df_S.to_csv(
"C:/Users/ravel/OneDrive/Área de Trabalho/DataScientist/sklearn/COVID-19/CasosPorEstado/Richards_covid19/data/sensibilidade/"
+ FILE,
sep=";")
binf = np.array([min(df_W.r), min(df_W.p), min(df_W.K), min(df_W.alfa)])
bsup = np.array([max(df_W.r), max(df_W.p), max(df_W.K), max(df_W.alfa)])
NSE_max = max(df_W.NSE)
for i in range(len(df_W)):
if df_W.NSE[i] == NSE_max:
pos_max = i
p0 = np.array([
df_W.r[pos_max], df_W.p[pos_max], df_W.K[pos_max], df_W.alfa[pos_max]
])
par = np.array([binf, bsup, p0])
df_par = pd.DataFrame(par, ["binf", "bsup", "p0"]).transpose()
df_par.to_csv(
"C:/Users/ravel/OneDrive/Área de Trabalho/DataScientist/sklearn/COVID-19/CasosPorEstado/Richards_covid19/data/range/"
+ FILE,
sep=";")
def Ajust_(FILE,pop,extrapolação,day_0,variavel,pasta, path_out):
data_covid = pd.read_csv(pasta+"/"+FILE, header = 0, sep = ";")
data_covid = data_covid[['DateRep',variavel,"cum-Deaths"]]
nome_data = 'DateRep'
df = data_covid
day_0_str=df[nome_data][0][:4]+'-'+df[nome_data][0][5:7]+'-'+df[nome_data][0][-2:]
#day_0_str = data_covid['DateRep'][0][-4:]+'-'+data_covid['DateRep'][0][3:5]+'-'+data_covid['DateRep'][0][:2]
date = np.array(day_0_str, dtype=np.datetime64)+ np.arange(len(data_covid))
if date[0]>=np.array(day_0, dtype=np.datetime64):
t0 = 0
else:
dif_dias =np.array(day_0, dtype=np.datetime64)-date[0]
t0 = dif_dias.astype(int)
date= date[t0:]
cumdata_covid = data_covid[['Cases']].cumsum()
cum_deaths = data_covid[['cum-Deaths']]
cumdata_cases = cumdata_covid['Cases'].values[t0:]
cum_deaths = cum_deaths['cum-Deaths'].values[t0:]
t = np.linspace(1,len(cumdata_cases),len(cumdata_cases))
N = pop*10**6
#n1,r1,beta,gama1,N,So,Uo,Qo,Co,Ro
n1_0,r1_0,n2_0,r2_0,beta_0,gama1_0,gama2_0,eta_0=[1/100,10,1/50,-4,.15,.05,.15,5/100]
So,Uo,Qo,Co,R1o,R2o,nCo = [.99*N,6*cumdata_cases[0],cumdata_cases[0],cumdata_cases[0],-4,14,3*cumdata_cases[0]]
p0 = [n1_0,r1_0,n2_0,r2_0,beta_0,gama1_0,gama2_0,eta_0,N,So,Uo,Qo,Co,R1o,R2o,3,nCo]
bsup = [n1_0*1.01,r1_0*1.1,n2_0*1.1, 0,0.50,.300,0.50, 20/100,N + 1, N,Uo*2.,Qo*2.0,Co+10**-9, 0,18,6,nCo*2.0]
binf = [n1_0*0.99,r1_0*0.9,n2_0*0.9,-5,0.01,.001,0.05,.01/100,N - 1,.9*N,Uo*.5,Qo*0.5,Co-10**-9,-6, 8,4,nCo*0.5]
# popt,pcov = ajust_curvefit(t,cum_deaths,p0,bsup,binf)
# perr = np.sqrt(np.diag(pcov))
# Nstd = nstf(.99)
# bsup = popt + Nstd * perr
# binf = popt - Nstd * perr
popt = min_minimize(cum_deaths,cumdata_cases,sucq_solve,p0,t,bsup,binf)
n1_0,r1_0,n2_0,r2_0,beta_0,gama1_0,gama2_0,eta_0,N,So,Uo,Qo,Co,R1o,R2o,Ro,nCo = popt
solution = SUCQ(t,*popt)
NSE = hy.nse(solution[:,3],cumdata_cases)
RMSE = hy.rmse(solution[:,3],cumdata_cases)
MARE = hy.mare(solution[:,3],cumdata_cases)
NSE_deaths = hy.nse(solution[:,7],cum_deaths)
RMSE_deaths = hy.rmse(solution[:,7],cum_deaths)
MARE_deaths = hy.mare(solution[:,7],cum_deaths)
print(FILE[9:-4])
print("n1 = %f " % (n1_0))
print("n2 = %f " % (n2_0))
print("r1m = %f" %(r1_0))
print("r2m = %f" %(r2_0))
print("R1o = %f" %(R1o))
print("R2o = %f" %(R2o))
print("beta = %f " % (beta_0))
print("gamma1 = %f " % (gama1_0))
print("gamma2 = %f " % (gama2_0))
print("eta = %f " % (eta_0))
print("NSE = %.5f " % (NSE))
print("NSE Deaths = %.5f " % (NSE_deaths))
print("#######################")
date_future = np.array(date[0], dtype=np.datetime64)+ np.arange(len(date)+extrapolação)
days_future = np.linspace(1,len(cumdata_cases)+extrapolação,len(cumdata_cases)+extrapolação)
Cum_cases_estimated = SUCQ(days_future, *popt)
estimativafutura_saída=pd.DataFrame(Cum_cases_estimated[:,3], columns = ["Cases"])
estimativafutura_saída["S"]=Cum_cases_estimated[:,0]
estimativafutura_saída["U"]=Cum_cases_estimated[:,1]
estimativafutura_saída["Q"]=Cum_cases_estimated[:,2]
estimativafutura_saída["I"] = Cum_cases_estimated[:,1]+Cum_cases_estimated[:,2]+Cum_cases_estimated[:,3]+Cum_cases_estimated[:,8]
estimativafutura_saída["uC"] =Cum_cases_estimated[:,8]
estimativafutura_saída["Rt"]=Cum_cases_estimated[:,6]
# R = []
# for i in days_future:
# R.append(R_s(i,n1_0,r1_0,n2_0,r2_0,R1o,R2o))
#
#estimativafutura_saída["Rt"]= np.array(R)
estimativafutura_saída["M"]=Cum_cases_estimated[:,7]
estimativafutura_saída["date"] = date_future
estimativafutura_saída.to_csv(path_out+'/'+FILE,sep=";")
return [n1_0,r1_0,n2_0,r2_0,beta_0,gama1_0,gama2_0,eta_0,N,So,Uo,Qo,Co,R1o,R2o,nCo, NSE, RMSE, MARE,NSE_deaths, RMSE_deaths, MARE_deaths]
def Ajust_SUCQ(FILE, pop, extrapolação):
path = 'C:/Users/ravellys/Documents/GitHub/COVID-19-Brasil/COVID-19-Brasil/data/DADOS/'
data_covid = pd.read_csv(path + FILE, header=0, sep=";")
data_covid = data_covid[['DateRep', 'Cases']]
nome_data = 'DateRep'
df = data_covid[['DateRep', 'Cases']]
day_0_str = df[nome_data][0][:4] + '-' + df[nome_data][0][5:7] + '-' + df[
nome_data][0][-2:]
day_0 = np.array(day_0_str, dtype=np.datetime64)
date = np.array(day_0_str, dtype=np.datetime64) + np.arange(
len(data_covid))
if date[0] >= np.array('2020-02-28', dtype=np.datetime64):
t0 = 0
else:
dif_dias = np.array('2020-02-28', dtype=np.datetime64) - date[0]
t0 = dif_dias.astype(int)
day_isol = '2020-03-20'
dif_dias = date[len(date) - 1] - np.array(day_isol, dtype=np.datetime64)
tf = dif_dias.astype(int)
if (day_0 <= np.array(day_isol, dtype=np.datetime64)):
date = date[t0:-tf]
cumdata_covid = data_covid[['Cases']].cumsum()
cumdata_cases = cumdata_covid['Cases'].values[t0:-tf]
days_mens = np.linspace(1, len(cumdata_cases), len(cumdata_cases))
N = pop * 10**6
t = days_mens
alfa_0, beta_0, gama1_0 = [.8 / N, .1, .19]
So, Uo, Qo, Co = [
.8 * N, 6 * cumdata_cases[0], cumdata_cases[0], cumdata_cases[0]
]
p0 = [alfa_0, beta_0, gama1_0, So, Uo, Qo, Co]
bsup = [1 / N, 1, 1, N, Uo * 1.5, Qo * 1.2, Co + 10**-9]
binf = [0, 0, 0, .5 * N, Uo * .5, Qo * 0.8, Co - 10**-9]
popt = ajust_curvefit(days_mens, cumdata_cases, p0, bsup, binf)
p0 = popt
popt = min_minimize(cumdata_cases, sucq_solve, p0, t, bsup, binf)
alfa_0, beta_0, gama1_0, So, Uo, Qo, Co = popt
solution = SUCQ(t, alfa_0, beta_0, gama1_0, So, Uo, Qo, Co)
NSE = hy.nse(cumdata_cases, solution[:, 3])
# print(FILE[9:-4])
# print("alfa = %f " % (alfa_0*N))
# print("beta = %f " % (beta_0))
# print("gamma = %f " % (gama1_0))
# print("R = %f" %(alfa_0*N/gama1_0))
# print("NSE = %.5f " % (NSE))
# print("#######################")
date_future = np.array(
date[0],
dtype=np.datetime64) + np.arange(len(date) + extrapolação)
days_future = np.linspace(1,
len(cumdata_cases) + extrapolação,
len(cumdata_cases) + extrapolação)
Cum_cases_estimated = SUCQ(days_future, *popt)
estimativafutura_saída = pd.DataFrame(Cum_cases_estimated[:, 3],
columns=["Cases"])
estimativafutura_saída["S"] = Cum_cases_estimated[:, 0]
estimativafutura_saída["U"] = Cum_cases_estimated[:, 1]
estimativafutura_saída["Q"] = Cum_cases_estimated[:, 2]
estimativafutura_saída[
"I"] = Cum_cases_estimated[:,
1] + Cum_cases_estimated[:,
2] + Cum_cases_estimated[:,
3]
estimativafutura_saída["date"] = date_future
fileout = 'C:/Users/ravellys/Documents/GitHub/COVID-19-Brasil/COVID-19-Brasil/data/DADOS_estimados_sem_isolamento/'
estimativafutura_saída.to_csv(fileout + FILE, sep=";")
if NSE >= 0.9:
return [alfa_0 * N / gama1_0, FILE[9:-4]]
c='green',
markersize=2,
markerfacecolor="None",
label='train')
plt.plot(x_test[:, len(x_test[1, :]) - 1],
y_test[:, i],
'o',
c='red',
markersize=2,
markerfacecolor="None",
label='test')
plt.plot(dias, TH[:, i], 'gray', label=r'$Simulado \theta\ %d$' % (i + 1))
NSEtest = he.nse(
TH[int(x_test[:, len(x_test[1, :]) - 1][0] - init - 1):, i], y_test[:,
i])
plt.text(260, .21, r"$NSE_{test}: %.3f $" % NSEtest, fontsize=12)
NSEtrain = he.nse(
TH[1:int(x_test[:, len(x_test[1, :]) - 1][0] - init - 1), i],
y_train[:, i])
plt.text(260, .23, r"$NSE_{train}: %.3f $" % NSEtrain, fontsize=12)
plt.xlabel('$dias$')
plt.ylabel(r'$ \theta\ (cm³.cm^{-3})$')
plt.ylim(.01, .25)
plt.legend(loc='upper left')
plt.show()
"""
Actual Evapotranspiration (ETa)