-
Notifications
You must be signed in to change notification settings - Fork 22
/
utils.R
143 lines (116 loc) · 3.46 KB
/
utils.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
library(forecast)
# Alternative to tsCV, to use only subset of last time_steps
# Warning: should be generalized
my_CV = function(y,
forecastfunction,
h=12,
last_year_months=NULL,
freq = 5,
start_year = 1997,
...){
if(is.null(last_year_months)){
i = 1
j = 1
last_year_months = list()
for(year in start_year:2010) {
for(month in 1:12){
if(i%%freq==0){
if ( (year!=2010) | (month <3) ){
last_year_months[[j]] <- c(year,month)
j = j+1
}
}
i = i+1
}
}
}
CV_residuals = NULL
for(last_year_month in last_year_months){
predictions = forecastfunction(window(y,end = last_year_month),h=h,...)$mean
true_values = window(y,start = last_year_month)[2:(h+1)]
CV_residuals = rbind(CV_residuals,true_values-predictions)
}
rownames(CV_residuals) = last_year_months
colnames(CV_residuals) = 1:h
return(CV_residuals)
}
# Some utility functions
my_forecast_plot = function(model,model_name){
fit = model(window(train_data, end=2007),h=h)
autoplot(window(train_data, start = 2004,end=2008)) +
autolayer(fit, PI=TRUE,alpha = 0.25,color = "red") +
autolayer(fit, PI=FALSE, series=model_name,size = 1) +
xlab("Year") + ylab("New orders index") +
ggtitle("Electrical equipment manufacturing (Euro area)") +
guides(colour=guide_legend(title="Forecast"))
}
print_scores = function(models_mae){
for (model_name in unique(models_mae$model)){
print(paste(model_name,":",round(mean(models_mae[models_mae$model==model_name,"mae"]),2)))
}
}
plot_scores = function(models_mae,naive = TRUE){
if(naive){
ggplot(models_mae,aes(x=horizon,y=mae,factor=model,color = model))+
geom_point()+
geom_line()+
scale_x_continuous(breaks = 1:12)
}else{
ggplot(models_mae[(models_mae$model != "Naïve") & (models_mae$model != "Seasonal Naïve"),],aes(x=horizon,y=mae,factor=model,color = model))+
geom_point()+
geom_line()+
scale_x_continuous(breaks = 1:12)
}
}
next_time_step = function(y, is_end = F){
if(is_end){
new_start = y
}else{
new_start = end(y)
}
if(new_start[2]==12){
new_start[1]=new_start[1]+1
new_start[2]=1
}else{
new_start[2]=new_start[2]+1
}
return(new_start)
}
previous_time_step = function(y){
new_start = end(y)
if(new_start[2]==1){
new_start[1]=new_start[1]-1
new_start[2]=12
}else{
new_start[2]=new_start[2]-1
}
return(new_start)
}
get_best_models = function(models_mae){
best_models = NULL
for(i in 1:12){
tmp = models_mae[models_mae$horizon==i,]
best_models = rbind(best_models,tmp[tmp$mae==min(tmp$mae),])
}
rownames(best_models)=NULL
return(best_models)
}
evaluate_test_residuals = function(fitted_model, model_function,data,...){
errors = NULL
for(i in 1:35){
data[1:(length(train_data)-12+i)] %>%
ts(frequency = 12,start = start(data)) %>%
model_function(model = fitted_model) %>% # use the fitted model to forecast test data
forecast(h=h,...) -> pred
true_values = c(window(data,start = start(pred$mean),end=end(pred$mean)))
pred_mean = c(pred$mean)
if(i < 12){
pred_mean[1:(12-i)]=NA
}
if(length(true_values) < 12){
true_values = c(true_values, rep(NA,12-length(true_values)))
}
errors = rbind(errors,true_values-pred_mean)
}
return(errors)
}