-
Notifications
You must be signed in to change notification settings - Fork 1
/
train_highprob_class.R
155 lines (128 loc) · 5.96 KB
/
train_highprob_class.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
##########################################################################
# Luz Frias, 2016-06-12
# Train the model
##########################################################################
library(data.table)
library(lubridate)
library(randomForest)
library(ggplot2)
library(caret)
source("R/air-data.R")
source("R/meteo-data.R")
source("R/traffic-data.R")
source("R/calendar-data.R")
# Merge air, variable, meteo and traffic data
air.data <- load_historical_air_data()
var.data <- load_contamination_variables()
meteo.data <- load_meteo_data()
#traffic.data <- load_traffic_data()
#calendar.data <- load_calendar_data()
air.data <- merge(air.data, var.data, by.x = "variable", by.y = "var_id")
air.data <- merge(air.data, meteo.data, by = c("date","hour"))
#air.data <- merge(air.data, traffic.data, by = "date")
#air.data <- merge(air.data, calendar.data, by = "date")
setkey(air.data, var_formula, station, date, hour)
# Feature normalization
#air.data[, m30_density := m30_density / 1e6]
prep.data <- air.data[valid == TRUE & var_formula == "NO2"]
# Feature generation
prep.data[, wday := as.factor(wday(date))]
prep.data[, month := as.factor(month(date))]
prep.data[, temp_range := max_temp - min_temp]
prep.data[, q_95 := quantile(value, .95), by = c("var_formula", "station", "date")]
prep.data[, prewarning := value > 180 & shift(value, type = "lag") > 180, by = c("var_formula", "station")]
prep.data[, warning := value > 200 & shift(value, type = "lag") > 200, by = c("var_formula", "station")]
# Aggregate by day
daily.data <- prep.data[, c("date", "var_formula", "station", "temp_range", "mean_temp", "min_temp", "month",
"max_temp","wind_speed","rel_humidity_pct", "wday", "rain", "q_95"),
with = FALSE]
setkey(daily.data, var_formula, station, date)
daily.data <- unique(daily.data)
danger.days <- prep.data[, .(prewarning = any(prewarning),
warning = any(warning)),
by = c("var_formula", "station", "date")]
daily.data <- merge(daily.data, danger.days, by = c("var_formula", "station", "date"))
setkey(daily.data, var_formula, station, date)
daily.data[, prev_q_95 := shift(q_95, type = "lag"), by = c("var_formula", "station")]
daily.data <- daily.data[!is.na(prev_q_95)]
# Feature selection
daily.data[, formula_station := paste(var_formula, station, sep = "_")]
daily.data <- daily.data[, c("date", "formula_station", "temp_range", "mean_temp", "min_temp", "rain", "month",
"max_temp","wind_speed", "rel_humidity_pct", "wday", "prewarning", "warning",
"prev_q_95"),
with = FALSE]
# Train a model for every station
variables <- c(quote(prewarning), quote(warning))
for (col in variables) {
probs <- data.frame(formula_station = character(0), prob = numeric(0),
precision = numeric(0), recall = numeric(0))
for (form_station in unique(daily.data[date > as.Date("2014-01-07")]$formula_station)) {
print(paste0("Training ", as.character(col), " for ", form_station))
data <- daily.data[formula_station == form_station ]
data[, formula_station := NULL]
data[, prewarning := as.factor(as.numeric(prewarning))]
data[, warning := as.factor(as.numeric(warning))]
# Train and test sets
train <- data[date < as.Date("2014-01-07")]
test <- data[date >= as.Date("2014-01-07")]
# Prepare for undersampling to balance the classes
train.p.pos <- train[eval(col) == "1"]
train.p.neg <- train[eval(col) == "0"]
if (nrow(train.p.pos) == 0) {
next
}
models <- 25
ntrees <- 750
for (i in 1:models) {
set.seed(1234+i)
print(paste0("model ", i))
train.p.neg.bal <- train.p.neg[sample(1:nrow(train.p.neg), nrow(train.p.pos) * 2),]
train.data <- rbindlist(list(train.p.pos, train.p.neg.bal))
train.data <- train.data[sample(nrow(train.data))]
train.class <- train.data[,eval(col)]
train.data[, date := NULL]
train.data[, prewarning := NULL]
train.data[, warning := NULL]
rf.model.tmp <- randomForest(train.data, y = train.class, method = "class",
ntree = ntrees, do.trace = FALSE)
png(paste0("res/high_prob_", as.character(col), "_imp_", i, "-", models, ".png"))
varImpPlot(rf.model.tmp)
dev.off()
png(paste0("res/high_prob_", as.character(col), "_error_", i, "-", models, ".png"))
plot(rf.model.tmp)
dev.off()
if (i == 1) {
rf.model <- rf.model.tmp
} else {
rf.model <- combine(rf.model.tmp, rf.model)
}
}
# Prediction
pred.rrff <-predict(rf.model, test, type = "prob")
# Plot real vs predicted values
png(filename = paste0("res/high_prob_", as.character(col), "_pred_", form_station, ".png"))
plot(test[,eval(col)], pred.rrff[,2])
dev.off()
# Confusion matrix
prob.rrff <- pred.rrff[,2]
calculate.cost <- function(prob) {
fp <- test[eval(col) == "0" & prob.rrff >= prob]
fn <- test[eval(col) == "1" & prob.rrff < prob]
nrow(fp) + nrow(fn) * 4
}
to.evaluate <- seq(.5,.95,.05)
costs <- sapply(to.evaluate, calculate.cost)
prob <- to.evaluate[which.min(costs)]
print(table(test[,eval(col)], pred.rrff[,2] > prob))
tp <- nrow(test[eval(col) == "1" & prob.rrff >= prob])
fp <- nrow(test[eval(col) == "0" & prob.rrff >= prob])
fn <- nrow(test[eval(col) == "1" & prob.rrff < prob])
# Save the model
save(rf.model, file = paste0("res/high_prob_", as.character(col), "_model_", form_station, ".model"))
# and the limit probability
probs <- rbind(probs, data.frame(formula_station = form_station, prob = prob,
precision = tp / (tp + fp), recall = tp / (tp + fn)))
}
write.table(probs, file = paste0("res/high_prob_", as.character(col), "_probs.tsv"),
row.names = FALSE, sep = "\t")
}