-
Notifications
You must be signed in to change notification settings - Fork 0
/
Variable_selection.R
220 lines (195 loc) · 13.1 KB
/
Variable_selection.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
rm(list=objects())
library(lubridate)
library(dplyr)
library(glmnet)
library(ranger)
library(Boruta)
library(VSURF)
setwd("~/Desktop/Simulations Covid Article/data")
set.seed(42)
### - Load Data
Data <- readRDS("Data_GAM.RDS")
Data <- filter(Data, Date >= ymd("2019-09-01"), Date < ymd("2020-09-17"), weights > 0)
Data <- filter(Data, region == 'France')
# Define usefull variables
lockdown_start <- ymd("2020-03-16")
### We select 4, 9, 14 variables using Lasso for each region, and use themm in the formula with the variable "WeekDays"
GAM_terms <- paste0("GAM_terms", 1:9)[-c(4,5)] # we remove terms 4 et 5, corresponding to toy and date
google_terms <- names(Data)[23:28]
complete_formula <- as.formula(paste0("residuals_norm ~ residuals_norm.48 + residuals_norm.336 + GovernmentResponseTracker + ",
paste0(GAM_terms, collapse = ' + '), " + ",
paste0(google_terms, collapse = ' + ')))
variables <- all.vars(complete_formula)[-1]
n_var <- length(variables)
formulas <- list() # formulas[[x]]stores the formula with c(5,10,15)[x] variables for regions[r]
# Initialise the formulas used for the RF using the two weeks before the start of the lockdown
sel <- which(Data$Date < lockdown_start & Data$Date >= ymd("2020-03-01"))
lasso <- glmnet(x = as.matrix(Data[sel, variables]), y = Data[sel, 'residuals_norm'], alpha = 1)
if (max(lasso$df) >= 4){
lambda.5 <- lasso$lambda[min(which((lasso$df >= 4)))]
}else{
lambda.5 <- lasso$lambda[length(lasso$df)]
}
vars.5 <- variables[which(glmnet(x = as.matrix(Data[sel, variables]), y = Data[sel, 'residuals_norm'], alpha = 1, lambda = lambda.5)$beta != 0)]
formulas[[1]] <- paste0("residuals_norm ~ WeekDays + ", paste0(vars.5, collapse = ' + '))
if (max(lasso$df) >= 9){
lambda.10 <- lasso$lambda[min(which((lasso$df >= 9)))]
}else{
lambda.10 <- lasso$lambda[length(lasso$df)]
}
vars.10 <- variables[which(glmnet(x = as.matrix(Data[sel, variables]), y = Data[sel, 'residuals_norm'], alpha = 1, lambda = lambda.10)$beta != 0)]
formulas[[2]] <- paste0("residuals_norm ~ WeekDays + ", paste0(vars.10, collapse = ' + '))
formulas[[3]] <- as.formula(paste0("residuals_norm ~ WeekDays + residuals_norm.48 + residuals_norm.336 + GovernmentResponseTracker + ",
paste0(GAM_terms, collapse = ' + '), " + ",
paste0(google_terms, collapse = ' + ')))
formulas[[4]] <- paste0("residuals_norm ~ ", paste0(getSelectedAttributes(Boruta(formulas[[3]], data=Data[sel,])), collapse = ' + '))
print(formulas)
# Initialise the results
forecast_date <- lockdown_start
results_GAM_RF_5 <- data.frame(Date = Date(0), region = character(0), tod = numeric(0), Load_norm = numeric(0), meanLoad = numeric(0),
GAM = numeric(0), residuals = numeric(0), q0.5 = numeric(0))
results_GAM_RF_10 <- data.frame(Date = Date(0), region = character(0), tod = numeric(0), Load_norm = numeric(0), meanLoad = numeric(0),
GAM = numeric(0), residuals = numeric(0),q0.5 = numeric(0))
results_GAM_RF_tot <- data.frame(Date = Date(0), region = character(0), tod = numeric(0), Load_norm = numeric(0), meanLoad = numeric(0),
GAM = numeric(0), residuals = numeric(0),q0.5 = numeric(0))
results_GAM_RF_Boruta <- data.frame(Date = Date(0), region = character(0), tod = numeric(0), Load_norm = numeric(0), meanLoad = numeric(0),
GAM = numeric(0), residuals = numeric(0), q0.5 = numeric(0))
sel_test <- which(Data$DateD == lockdown_start)
results_GAM_RF_5 <- rbind(results_GAM_RF_5,
data.frame(Date = Data$Date[sel_test], region = Data$region[sel_test], tod = Data$tod[sel_test],
Load_norm = Data$Load_norm[sel_test], meanLoad = Data$meanLoad[sel_test],
GAM = Data$pred_norm[sel_test], residual = Data$residuals_norm[sel_test],
q0.5 = 0))
results_GAM_RF_10 <- rbind(results_GAM_RF_10,
data.frame(Date = Data$Date[sel_test], region = Data$region[sel_test], tod = Data$tod[sel_test],
Load_norm = Data$Load_norm[sel_test], meanLoad = Data$meanLoad[sel_test],
GAM = Data$pred_norm[sel_test], residual = Data$residuals_norm[sel_test],
q0.5 = 0))
results_GAM_RF_tot <- rbind(results_GAM_RF_tot,
data.frame(Date = Data$Date[sel_test], region = Data$region[sel_test], tod = Data$tod[sel_test],
Load_norm = Data$Load_norm[sel_test], meanLoad = Data$meanLoad[sel_test],
GAM = Data$pred_norm[sel_test], residual = Data$residuals_norm[sel_test],
q0.5 = 0))
results_GAM_RF_Boruta <- rbind(results_GAM_RF_Boruta,
data.frame(Date = Data$Date[sel_test], region = Data$region[sel_test], tod = Data$tod[sel_test],
Load_norm = Data$Load_norm[sel_test], meanLoad = Data$meanLoad[sel_test],
GAM = Data$pred_norm[sel_test], residual = Data$residuals_norm[sel_test],
q0.5 = 0))
dates <- unique(filter(Data, Date > lockdown_start + duration(1, units = "days"))$DateD)
N <- length(dates)
## %%%%%%%%%%%%% Compute the predictions for the pandemix period %%%%%%%%%%%%%%% ##
for (j in 1:N){
forecast_date <- dates[j]
if (j %% 7 == 0){ # update formulas every week
sel <- which(Data$Date >= lockdown_start & Data$Date < forecast_date)
lasso <- glmnet(x = as.matrix(Data[sel, variables]), y = Data[sel, 'residuals_norm'], alpha = 1)
if (max(lasso$df) >= 4){
lambda.5 <- lasso$lambda[min(which((lasso$df >= 4)))]
}else{
lambda.5 <- lasso$lambda[length(lasso$df)]
}
vars.5 <- variables[which(glmnet(x = as.matrix(Data[sel, variables]), y = Data[sel, 'residuals_norm'], alpha = 1, lambda = lambda.5)$beta != 0)]
formulas[[1]]<- paste0("residuals_norm ~ WeekDays + ", paste0(vars.5, collapse = ' + '))
if (max(lasso$df) >= 9){
lambda.10 <- lasso$lambda[min(which((lasso$df >= 9)))]
}else{
lambda.10 <- lasso$lambda[length(lasso$df)]
}
vars.10 <- variables[which(glmnet(x = as.matrix(Data[sel, variables]), y = Data[sel, 'residuals_norm'], alpha = 1, lambda = lambda.10)$beta != 0)]
formulas[[2]]<- paste0("residuals_norm ~ WeekDays + ", paste0(vars.10, collapse = ' + '))
vars.Boruta <- getSelectedAttributes(Boruta(formulas[[3]], data=Data[sel,]))
formulas[[4]]<- as.formula(paste0("residuals_norm ~ ", paste0(vars.Boruta, collapse = ' + ')))
print(forecast_date)
print(formulas[[4]])
}
# Train the RF using the avalaible pandemic data, produce forecasts
for (i in 0:47){
sel_test <- which(Data$tod == i & Data$DateD == forecast_date)
if (length(sel_test) != 0){
sel_train <- which(Data$tod == i & Data$Date >= lockdown_start & Data$Date < forecast_date)
rf.5 <- ranger(formulas[[1]], data = Data[sel_train,], quantreg = T)
rf.5.quantiles <- predict(rf.5, type = "quantiles", data = Data[sel_test,], quantiles = c(0.5, 0.1, 0.5, 0.9, 0.95))$predictions
results_GAM_RF_5 <- rbind(results_GAM_RF_5,
data.frame(Date = Data$Date[sel_test], region = Data$region[sel_test], tod = Data$tod[sel_test],
Load_norm = Data$Load_norm[sel_test], meanLoad = Data$meanLoad[sel_test],
GAM = Data$pred_norm[sel_test], residual = Data$residuals_norm[sel_test],
q0.5 = rf.5.quantiles[1]))
rf.10 <- ranger(formulas[[2]], data = Data[sel_train,], quantreg = T)
rf.10.quantiles <- predict(rf.10, type = "quantiles", data = Data[sel_test,], quantiles = c(0.5))$predictions
results_GAM_RF_10 <- rbind(results_GAM_RF_10,
data.frame(Date = Data$Date[sel_test], region = Data$region[sel_test], tod = Data$tod[sel_test],
Load_norm = Data$Load_norm[sel_test], meanLoad = Data$meanLoad[sel_test],
GAM = Data$pred_norm[sel_test], residual = Data$residuals_norm[sel_test],
q0.5 = rf.10.quantiles[1]))
rf.tot <- ranger(formulas[[3]], data = Data[sel_train,], quantreg = T)
rf.tot.quantiles <- predict(rf.tot, type = "quantiles", data = Data[sel_test,], quantiles = c(0.5, 0.1, 0.5, 0.9, 0.95))$predictions
results_GAM_RF_tot <- rbind(results_GAM_RF_tot,
data.frame(Date = Data$Date[sel_test], region = Data$region[sel_test], tod = Data$tod[sel_test],
Load_norm = Data$Load_norm[sel_test], meanLoad = Data$meanLoad[sel_test],
GAM = Data$pred_norm[sel_test], residual = Data$residuals_norm[sel_test],
q0.5 = rf.tot.quantiles[1]))
rf.Boruta <- ranger(formulas[[4]], data = Data[sel_train,], quantreg = T)
rf.Boruta.quantiles <- predict(rf.Boruta, type = "quantiles", data = Data[sel_test,], quantiles = c(0.5, 0.1, 0.5, 0.9, 0.95))$predictions
results_GAM_RF_Boruta <- rbind(results_GAM_RF_Boruta,
data.frame(Date = Data$Date[sel_test], region = Data$region[sel_test], tod = Data$tod[sel_test],
Load_norm = Data$Load_norm[sel_test], meanLoad = Data$meanLoad[sel_test],
GAM = Data$pred_norm[sel_test], residual = Data$residuals_norm[sel_test],
q0.5 = rf.Boruta.quantiles[1]))
}
}
}
# Save results
saveRDS(results_GAM_RF_5, 'results_VarSel_5.RDS')
saveRDS(results_GAM_RF_10, 'results_VarSel_10.RDS')
saveRDS(results_GAM_RF_tot, 'results_VarSel_tot.RDS')
saveRDS(results_GAM_RF_Boruta, 'results_VarSel_Boruta.RDS')
## %%%%%%%%%%%%% Comare variable selection methods %%%%%%%%%%%%%%% ##
rm(list=objects())
library(lubridate)
library(dplyr)
library(opera)
setwd("~/Desktop/Simulations Covid Article/data")
# Agregate results of the individual quantile RF for different number of variables region by region, time of day by time of day and quantile by quantile
results_5 <- readRDS('results_VarSel_5.RDS')
results_10 <- readRDS('results_VarSel_10.RDS')
results_tot <- readRDS('results_VarSel_tot.RDS')
results_Boruta <- readRDS('results_VarSel_Boruta.RDS')
results_agg <- data.frame(Date = Date(0), tod = numeric(0),
meanLoad = numeric(0), Load_norm = numeric(0),
GAM = numeric(0),
q0.05 = numeric(0), q0.1 = numeric(0), q0.5 = numeric(0),
q0.9 = numeric(0), q0.95 = numeric(0), mean = numeric(0))
N <- length(which(results_5$tod == 0))
Var_sel_weights <- matrix(nrow = N, ncol = 3, 0)
pred_i <- data.frame(Date = Date(0), tod = numeric(0),
region = character(0), meanLoad = numeric(0),
Load_norm = numeric(0), GAM = numeric(0),
GAM_Lasso = numeric(0), GAM_tot = numeric(0), GAM_Boruta = numeric(0))
for (i in 0:47){
sel_i<- which(results_5$tod == i)
sel_i_Boruta <- which(results_Boruta$tod == i)
experts <- as.matrix(cbind(results_5[sel_i, "q0.5"],
results_10[sel_i, "q0.5"],
results_tot[sel_i, "q0.5"]))
colnames(experts) <- c("5var", "10var", "tot")
Y <- results_5[sel_i, "residual"]
agg <- opera::mixture(Y = Y, experts = experts, model = "MLpol", loss.gradient = TRUE, loss.type = c(name = 'pinball', tau = 0.5))
pred_i <- rbind(pred_i,
data.frame(Date = ymd(floor_date(results_5$Date[sel_i], unit = 'day')), tod = i,
region = results_5$region[sel_i], meanLoad = results_5$meanLoad[sel_i],
Load_norm = results_5$Load_norm[sel_i], GAM = results_5$GAM[sel_i],
GAM_Lasso = results_5$GAM[sel_i] + agg$prediction,
GAM_tot = results_tot$GAM[sel_i] + results_tot$q0.5[sel_i],
GAM_Boruta = results_Boruta$GAM[sel_i] + results_Boruta$q0.5[sel_i]))
print(i)
}
res_1 <- filter(pred_i, Date <= ymd("2020-05-11"))
res_2 <- filter(pred_i, Date > ymd("2020-05-11"))
mape_1_lasso <- round(100*mean(abs(res_1$Load_norm - res_1$GAM_Lasso)/res_1$Load_norm),2)
mape_1_tot <- round(100*mean(abs(res_1$Load_norm - res_1$GAM_tot)/res_1$Load_norm),2)
mape_1_Boruta <- round(100*mean(abs(res_1$Load_norm - res_1$GAM_Boruta)/res_1$Load_norm),2)
mape_2_lasso <- round(100*mean(abs(res_2$Load_norm - res_2$GAM_Lasso)/res_2$Load_norm),2)
mape_2_tot <- round(100*mean(abs(res_2$Load_norm - res_2$GAM_tot)/res_2$Load_norm),2)
mape_2_Boruta <- round(100*mean(abs(res_2$Load_norm - res_2$GAM_Boruta)/res_2$Load_norm),2)
print(paste(mape_1_lasso, mape_1_Boruta, mape_1_tot))
print(paste(mape_2_lasso, mape_2_Boruta, mape_2_tot))