-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathres_train_21746332_180423_0.9730.R
172 lines (122 loc) · 5.66 KB
/
res_train_21746332_180423_0.9730.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
if (!require("pacman")) install.packages("pacman")
pacman::p_load(knitr, tidyverse, highcharter, data.table, lubridate, pROC, tictoc, DescTools, lightgbm)
set.seed(84)
options(scipen = 9999, warn = -1, digits= 4)
train <- fread("train_reduce_2.csv",
col.names =c("ip", "app", "device", "os", "channel", "click_time",
"is_attributed", "ip_nextClick", "ip_app_nextClick", "ip_channel_nextClick", "ip_os_nextClick"),
showProgress = FALSE)
invisible(gc())
most_freq_hours_in_test_data <- c("4","5","9","10","13","14")
least_freq_hours_in_test_data <- c("6","11","15")
train$ip_nextClick <- NULL
train$ip_os_nextClick <- NULL
train[, UsrappCount:=.N, by=list(ip,app,device,os)]
train[, app_f := .N, by = "app"]
train[, ip_dev_f := .N, by = "ip,device"]
train[, ip_os_f := .N, by = "ip,os"]
train[, ip_app_f := .N, by = "ip,app"]
train[, channel_f := .N, by = "channel"]
train <- train %>% mutate(wday = Weekday(click_time),
hour = hour(click_time),
in_test_hh = ifelse(hour %in% most_freq_hours_in_test_data, 1,
ifelse(hour %in% least_freq_hours_in_test_data, 3, 2)))
train <- train %>%
select(-c(click_time)) %>%
add_count(ip, wday, in_test_hh) %>% rename("nip_day_test_hh" = n) %>%
select(-c(in_test_hh)) %>%
add_count(ip) %>% rename("n_rowip" = n) %>%
add_count(os, device) %>% rename("n_os_dev" = n) %>%
add_count(os, app, channel) %>% rename("n_os_app_chan" = n) %>%
add_count(app, channel) %>% rename("n_app_chan" = n) %>%
add_count(ip, wday, hour) %>% rename("n_ip" = n) %>%
add_count(ip, wday, hour, os) %>% rename("n_ip_os" = n) %>%
add_count(ip, wday, hour, app) %>% rename("n_ip_app" = n) %>%
add_count(ip, wday, hour, app, os) %>% rename("n_ip_app_os" = n) %>%
add_count(app, wday, hour) %>% rename("n_app" = n) %>%
add_count(ip, device, wday, hour) %>% rename("nip_dev_d_h" = n) %>%
select(-c(wday)) %>% select(-c(ip))
gc()
train[is.na(train)] <- 0
gc()
library(caret)
set.seed(71)
train_part <- createDataPartition(train$is_attributed, p = 0.7, list = FALSE)
categorical_features = c("app", "os", "channel", "hour")
dtrain = lgb.Dataset(data.matrix(train[train_part,] %>% select(-is_attributed)),
label = as.numeric(train[train_part,]$is_attributed),
categorical_feature = categorical_features)
dvalid = lgb.Dataset(data.matrix(train[-train_part,] %>% select(-is_attributed)),
label = as.numeric(train[-train_part,]$is_attributed),
categorical_feature = categorical_features)
rm(train)
invisible(gc())
params = list(objective = "binary",
metric = "auc",
learning_rate= 0.1,
num_leaves= 7,
max_depth= 3,
min_child_samples= 100,
max_bin= 255, # RAM dependent as per LightGBM documentation
subsample= 0.7,
subsample_freq= 1,
colsample_bytree= 0.7,
min_child_weight= 0,
min_split_gain= 0,
scale_pos_weight=99.75) # calculated for this dataset
set.seed(71)
model <- lgb.train(params, dtrain, valids = list(validation = dvalid), nthread = 7,
nrounds = 3000, verbose= 1, early_stopping_rounds = 50, eval_freq = 25)
kable(lgb.importance(model, percentage = TRUE))
#Predict
test <- fread("test.csv", colClasses = list(numeric=2:6), showProgress = FALSE)
sub <- data.table(click_id = test$click_id, is_attributed = NA)
test$click_id <- NULL
invisible(gc())
test$ip_nextClick <- NULL
test$ip_os_nextClick <- NULL
test[, UsrappRank:=1:.N, by=list(ip,app,device,os)]
test[, UsrappCount:=.N, by=list(ip,app,device,os)]
test[, UsrappNewness:=1:.N, by=list(ip,app,device,os)]
test[, UsrCount:=.N, by=list(ip,device,os)]
test[, UsrNewness:=1:.N, by=list(ip,device,os)]
test[, app_f := .N, by = "app"]
test[, ip_dev_f := .N, by = "ip,device"]
test[, ip_os_f := .N, by = "ip,os"]
test[, ip_app_f := .N, by = "ip,app"]
test[, channel_f := .N, by = "channel"]
test <- test %>% mutate(wday = Weekday(click_time),
hour = hour(click_time),
in_test_hh = ifelse(hour %in% most_freq_hours_in_test_data, 1,
ifelse(hour %in% least_freq_hours_in_test_data, 3, 2)))
test <- test %>%
select(-c(click_time)) %>%
add_count(ip, wday, in_test_hh) %>% rename("nip_day_test_hh" = n) %>%
select(-c(in_test_hh)) %>%
add_count(ip) %>% rename("n_rowip" = n) %>%
add_count(os, device) %>% rename("n_os_dev" = n) %>%
add_count(os, app, channel) %>% rename("n_os_app_chan" = n) %>%
add_count(ip, device) %>% rename("n_ip_dev" = n) %>%
add_count(app, channel) %>% rename("n_app_chan" = n) %>%
add_count(ip, wday, hour) %>% rename("n_ip" = n) %>%
add_count(ip, wday, hour, os) %>% rename("n_ip_os" = n) %>%
add_count(ip, wday, hour, app) %>% rename("n_ip_app" = n) %>%
add_count(ip, wday, hour, app, os) %>% rename("n_ip_app_os" = n) %>%
add_count(app, wday, hour) %>% rename("n_app" = n) %>%
add_count(device, hour) %>% rename("ndev_h" = n) %>%
add_count(ip, device, wday, hour) %>% rename("nip_dev_d_h" = n) %>%
add_count(channel, hour, device) %>% rename("nchan_dev_h" = n) %>%
select(-c(wday)) %>% select(-c(ip))
gc()
test[is.na(test)] <- 0
gc()
preds <- predict(model, data = data.matrix(test[, colnames(test)]), n = model$best_iter)
preds <- as.data.frame(preds)
sub$is_attributed <- NULL
sub$is_attributed <- preds
rm(test)
invisible(gc())
fwrite(sub, "res_train_21746332_180423.csv")
summary(preds$preds)
kable(lgb.importance(model, percentage = TRUE))
#0.9730