-
Notifications
You must be signed in to change notification settings - Fork 0
/
script.R
447 lines (401 loc) · 16.1 KB
/
script.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
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
# Importing data
setwd("Downloads/shuuheialb.github.io/projects/employee-attrition")
rm(list = ls())
DATA <- read.csv("hr_data.csv")
# Save some test and validation data
library(rsample)
set.seed(100)
SPLIT <- initial_validation_split(DATA, c(0.75, 0.15))
TEST <- testing(SPLIT)
hr_valid <- validation(SPLIT)
hr <- training(SPLIT)
hr_full <- rbind(hr, hr_valid)
# Check for incorrect/problematic entries
# First few rows
#library(DataExplorer)
library(dplyr)
library(ggplot2)
library(reshape2)
#head(hr)
# All columns' types and freq/dist (summary and plots)
#str(hr)
#summary(hr)
#plot_bar(hr, title = "Categorical Variable Frequency", nrow = 2, ncol = 2)
#plot_histogram(hr, title = "Numerical Variable Distribution", nrow = 3, ncol = 3)
# Info about missing values and outliers
#plot_intro(hr, title = "Missing Values")
# Constant vectors:
# Formatted columns
cat_cols <- c("BusinessTravel", "Department", "Gender", "EducationField", "JobRole", "MaritalStatus")
bool_cols <- c("Attrition", "Over18", "OverTime")
char_to_bool <- function (col) { col %in% c("Yes", "Y", TRUE) }
# Irrelevant columns
is_single_col <- function (col) { length(unique(col)) == 1 }
invalid_unif_cols <- c("DailyRate", "MonthlyRate", "HourlyRate")
# Correlated columns
coupled_temporal_cols <- c("Age", "TotalWorkingYears", "YearsWithCurrManager", "YearsInCurrentRole", "YearsSinceLastPromotion")
high_corr_cols <- c("JobLevel")
library(survival)
library(randomForestSRC)
library(SurvMetrics)
library(pec)
# === Models
TIME <- "YearsAtCompany"
STATUS <- "Attrition"
ID <- "EmployeeNumber"
coxph_model <- function (df) {
formula <- as.formula(paste("Surv(", TIME, ", ", STATUS, ") ~ ."))
model <- coxph(data = df %>% select(-any_of(ID)), formula = formula, method = "efron", x = TRUE)
return(model)
}
rsf_model <- function (df, params = NULL) {
if (is.null(params)) {
params <- list(ntree = 100, mtry = NULL)
}
formula <- as.formula(paste("Surv(", TIME, ", ", STATUS, ") ~ ."))
model <- rfsrc(data = df %>% select(-any_of(ID)), formula = formula,
ntree = params$ntree, mtry = params$mtry)
}
model_f <- function (mode, params = NULL) { if (mode == "coxph") coxph_model
else if (mode == "rsf") function(df) { rsf_model(df, params = params) }
else NULL }
# === Feature selection
# 1. Cox score
coxph_score <- function (df, model) { # ignore the model
features <- names(df)[!(names(df) %in% c(TIME, STATUS))]
scores <- sapply(features, function (col){
summary(coxph_model(df[c(TIME, STATUS, col)]))$concordance[1]
})
names(scores) <- features
return(scores)
}
coxph_details <- function (model) {
return(summary(model)$coefficients[, c("exp(coef)", "coef", "se(coef)")])
}
# 2. RSF Importance
rsf_importance <- function (df, model) {
return(predict(model, data = df, importance = TRUE)$importance)
}
feature_rank_f <- function (mode) { if (mode == "coxph") coxph_score else if (mode == "rsf") rsf_importance else NULL }
# ===
print_feature_rank <- function (df, mode) {
if (mode == "coxph") {
print("Univariate Cox score:")
} else {
print("Variable importance:")
}
res <- feature_rank_f(mode)(df, model_f(mode)(df))
print(res[order(res, decreasing = TRUE)])
}
# === Accuracy metrics
# SurvMetrics variables incompatible with custom functions:
# model after df then tested after df2
survMetrics_wrapper <- function (df, df2, mode, metric_f) {
tmp <- list(time = TIME, status = STATUS)
TIME <<- "time"
STATUS <<- "status"
rename_survmetrics <- function (df) {
names(df)[names(df) == tmp$time] <- TIME
names(df)[names(df) == tmp$status] <- STATUS
return(df)
}
res <- metric_f(model_f(mode)(rename_survmetrics(df)), rename_survmetrics(df2), t_star = -1)
TIME <<- tmp$time
STATUS <<- tmp$status
return(res)
}
# 1. Concordance index
c_index <- function (df, df2, mode) {
return(survMetrics_wrapper(df, df2, mode, SurvMetrics::Cindex))
}
# 2. Brier score: SOON
brier_score <- function (df, df2, mode) {
return(survMetrics_wrapper(df, df2, mode, SurvMetrics::Brier))
}
# Legacy code from pec:
plot_pec <- function (df, model) { # split???
formula <- as.formula(paste("Surv(", TIME, ", ", STATUS, ") ~ ."))
pred_error <- pec(model, data = df, splitmethod = "cv10", formula = formula, cens.model = "marginal", reference = FALSE)
plot(pred_error, xlim = c(0, 10), ylim = c(0, 0.25)) # 0.25 is the worst case scenario (random model)
title("Prediction Error Curve")
}
# ===
print_c_index <- function (df, df2, mode) {
print("Concordance index:")
print(c_index(df, df2, mode))
}
# === Cross-validation
cross_val <- function (df, mode, feature_num = 1, params = NULL, k = 5) {
random_index <- sample(1:nrow(df))
performance_score1_vec <- numeric(k)
for (fold in 1:k) {
prev_end <- round((fold-1)/k * nrow(df))
size <- round(fold * nrow(df)/k) - round((fold-1) * nrow(df)/k)
idx <- random_index[(prev_end+1):(prev_end+size)]
test <- df[idx, ]
train <- df[-idx, ]
# Preprocessing
tt <- prepare(train, test, mode)
train <- tt[[1]]
test <- tt[[2]]
# Initial model
model1 <- model_f(mode, params)(train)
feature_rank1 <- feature_rank_f(mode)(test, model1)
feature_rank1 <- feature_rank1[order(feature_rank1, decreasing = TRUE)]
# Correlated columns removal
feature_redundancy <- corr_table(train, limit = 0.8)
for (i in 1:nrow(feature_redundancy)) {
# Retain higher rank, if any
x <- feature_redundancy[i, "Var1"]
y <- feature_redundancy[i, "Var2"]
if (x %in% feature_rank1 && y %in% feature_rank1) {
x_rank <- which(feature_rank1 == x)
y_rank <- which(feature_rank1 == y)
feature_rank1 <- feature_rank1[-max(x_rank, y_rank)]
}
}
# Next model
feature_rank2 <- names(feature_rank1)[1:min(feature_num, length(names(feature_rank1)))]
#print("Choosing features ...")
#print(feature_rank2)
train2 <- train[c(TIME, STATUS, feature_rank2)]
test2 <- test[c(TIME, STATUS, feature_rank2)]
performance_score1 <- c_index(train2, test2, mode)
# Brier score soon
performance_score1_vec[fold] <- performance_score1
}
# !!! Box plot soon
print(paste0("Performance score via concordance index, for feature number = ", feature_num, ":"))
print(paste0("Mean: ", mean(performance_score1_vec)))
print(performance_score1_vec)
print(paste0("SE: ", sd(performance_score1_vec)))
}
# === Preprocessing: very dataset-specific so no need for TIME/STATUS/ID here
prepare <- function (df, df2, mode = "") {
df <- prepare_gen(df)
df2 <- prepare_gen(df2)
# Train set additional process: intern and outlier removal
df <- df %>%
mutate(IsIntern = (MonthlyIncome < 2000)) %>%
mutate(iqr = IQR(YearsAtCompany),
q1 = quantile(YearsAtCompany, probs = c(0.25)) - 3*iqr,
q3 = quantile(YearsAtCompany, probs = c(0.75)) + 3*iqr,
IsOutlier = between(YearsAtCompany, q1, q3)) %>%
select(-c(iqr, q1, q3))
df_train <- df %>% filter(!(IsIntern & IsOutlier)) %>% select(-c(IsIntern, IsOutlier))
df_removed <- df %>% filter(IsIntern & IsOutlier) %>% select(-c(IsIntern, IsOutlier))
# Assumption check for CoxPH model
if (mode == "coxph") {
ph_obj <- df_train %>%
select(-c(Department, JobRole)) %>%
coxph_model %>% proportional_hazard_table
non_ph_cols <- names(which(ph_obj$table[, "p"] < 0.01))
non_ph_cols <- non_ph_cols[!(non_ph_cols == "GLOBAL")]
for (col in non_ph_cols) { # There should be a better Tidyverse syntax
#inter_col <- paste0("Inter_TIME_", col)
#df_train[inter_col] <- df_train[TIME] * df_train[col]
#df2[inter_col] <- df2[TIME] * df2[col]
#df_removed[inter_col] <- df_removed[TIME] * df_removed[col]
}
}
return(list(df_train, df2, df_removed))
}
prepare_gen <- function (df) {
df %>%
mutate_at(cat_cols, factor) %>%
mutate_at(bool_cols, char_to_bool) %>%
select(-where(is_single_col)) %>%
select(-any_of(invalid_unif_cols)) %>%
mutate(NotWorkingYears = Age - TotalWorkingYears) %>%
mutate(YearsAtOtherCompanies = TotalWorkingYears - YearsAtCompany) %>%
select(-any_of(coupled_temporal_cols)) %>%
select(-any_of(high_corr_cols))
}
proportional_hazard_table <- function (model) {
return(cox.zph(model))
}
corr_table <- function (df, limit = 0){
# Convert categories to dummies
df2 <- model.matrix(~ 0+., df)
# Dropping entries
corr <- cor(df2)
corr[lower.tri(corr, diag = TRUE)] <- NA
corr[abs(corr) == 1] <- NA
# Convert to long format, sorted
res <- melt(corr)
res <- na.omit(res)
res <- res[abs(res$value) > limit,]
res <- res[order(res$value, decreasing = TRUE),]
rownames(res) <- 1:nrow(res)
return(res)
}
# Final iteration
coxph_cols <- c("Attrition", "YearsAtCompany", "EmployeeNumber", "MonthlyIncome", "JobRole", "OverTime")
rsf_cols <- c("Attrition", "YearsAtCompany", "EmployeeNumber", "OverTime", "MonthlyIncome", "StockOptionLevel", "TrainingTimesLastYear",
"EnvironmentSatisfaction", "JobRole", "NumCompaniesWorked", "NotWorkingYears", "WorkLifeBalance")
TT_c <- prepare(hr_full, TEST, mode = "coxph")
TT_r <- prepare(hr_full, TEST, mode = "rsf")
TRAIN_c <- TT_c[[1]][coxph_cols]
TRAIN_r <- TT_r[[1]][rsf_cols]
coxph_final_model <- coxph_model(TRAIN_c)
rsf_final_model <- rsf_model(TRAIN_r)
# === Plotly Visualisation: Legacy Code
# Preparing the test data, rejoining training with ousted data
TEST_c <- TT_c[[2]][coxph_cols]
TEST_r <- TT_r[[2]][rsf_cols]
hr_full_c <- rbind(TRAIN_c, TT_c[[3]][coxph_cols])
hr_full_r <- rbind(TRAIN_r, TT_r[[3]][rsf_cols])
# Appending predictions
years = 5
rsf_risk <- predict(rsf_final_model, hr_full_r)$chf[, 1 + (0:years)] # matrix
rownames(rsf_risk) <- rownames(hr_full_r) # For Plotly
hr_full_r["Hazard_RSF_in_5yr"] <- rsf_risk[, 6]
hr_full_c["HazardRatio_CoxPH"] <- predict(coxph_final_model, hr_full_c, type = "risk")
# Table sorted by hazards
cutoff = 20
sorted <- function (df, key_col) {
df <- df[df["Attrition"] == FALSE, ]
return(df[order(df[, key_col], decreasing = TRUE), ][1:cutoff, ])
}
hr_sorted_rsf <- sorted(hr_full_r, "Hazard_RSF_in_5yr")
hr_sorted_coxph <- sorted(hr_full_c, "HazardRatio_CoxPH")
head(hr_sorted_rsf)
head(hr_sorted_coxph)
# Plotly implementation
suppressMessages(library(plotly))
fig <- plot_ly()
# --- Adding traces
# RSF
for (rank in 1:cutoff) {
employee_row <- hr_sorted_rsf[rank, ]
employee_row_idx_str <- rownames(employee_row)
fig <- fig %>% add_trace(x = 0:years,
y = rsf_risk[employee_row_idx_str, ],
name = employee_row$EmployeeNumber,
type = "scatter", mode = "lines+markers", # Every time you put type, it assign DOMNum = 0
marker = list(size = 6),
line = list(shape = "spline", width = 2),
hovertemplate = paste0("<b>", rank, ". Employee ", employee_row$EmployeeNumber, "</b>",
"<br>Year: %{x}",
"<br>Cumulative risk: %{y:.2f}"),
hoverlabel = list(font_size = 16),
legendrank = rank,
legendgroup = employee_row$EmployeeNumber,
showlegend = TRUE)
}
# CoxPH
fig <- fig %>% add_trace(data = hr_sorted_coxph,
x = 1:cutoff,
y = ~HazardRatio_CoxPH,
type = "bar",
marker = list(color = 1:cutoff, colorscale = list(c(0, "#64a1f4"), c(1, "#bfe6b5"))), # blue-green
hovertemplate = paste0("<b>%{x}. Employee ", hr_sorted_coxph$EmployeeNumber, "</b>",
"<br>Risk ratio: %{y:.2f}"),
hoverlabel = list(bgcolor = "white", font_size = 16),
showlegend = FALSE,
visible = FALSE)
# --- Decorators
# Including the baseline
avg_cum_hazard <- -log(1-mean(hr_full_r[, "Attrition"]))
baseline <- list(
type = "line", x0 = 0, x1 = 5, y0 = avg_cum_hazard, y1 = avg_cum_hazard,
line = list(dash = "dash", width = 4, color = "#82A0D8")
)
fig <- fig %>% add_text(
x = 0.3, y = 0.22,
text = "Average hazard", textfont = list(family = "sans serif", size = 12),
showlegend = FALSE
)
# Helper functions to group CHF based on variables
library(RColorBrewer)
unique_index <- function(x) {
unique_vals <- unique(x)
first_occur <- integer(length(unique_vals))
for (i in 1:length(unique_vals)) {
first_occur[i] <- which(x == unique_vals[i])[1]
}
return(first_occur)
}
create_buttons <- function(vars) {
lapply(vars, function(var) {
color_num <- 8
color_palletes <- brewer.pal(color_num, "Set2")
var_factors <- as.numeric(factor(hr_sorted_rsf[[var]]))
button <- list(
method = "restyle",
label = var,
args = list(list( # Please read Plotly.js for detailed documentation (not R/Python)
line.color = color_palletes[var_factors %% color_num + 1],
marker.color = color_palletes[var_factors %% color_num + 1],
legendrank = if(var == "EmployeeNumber") 1:cutoff else var_factors,
legendgroup = hr_sorted_rsf[[var]],
showlegend = 1:cutoff %in% unique_index(var_factors),
name = as.character(hr_sorted_rsf[[var]])
), 1:cutoff - 1)
)
return(button)
})
}
# --- Setting layout
rsf_layout = list(
title = "Top 20 High-Risk Employees, according to RSF's Cumulative Risk in Five Years",
shapes = list(baseline),
xaxis = list(title = "Year", range = c(0, 5), zerolinecolor = 'white'),
yaxis = list(title = "Accummulated Risk over Time", range = c(0, 1.2), zerolinecolor = 'white'),
`annotations[1].visible` = TRUE,
`updatemenus[1].visible` = TRUE
)
coxph_layout = list(
title = "Top 20 High-Risk Employees, according to CoxPH's Risk Ratio",
shapes = list(),
xaxis = list(title = "Employee ID", tickmode = "array", tickvals = 1:cutoff,
ticktext = hr_sorted_coxph$EmployeeNumber, zerolinecolor = 'white'),
yaxis = list(title = "Risk Ratio", zerolinecolor = 'white'),
`annotations[1].visible` = FALSE,
`updatemenus[1].visible` = FALSE
)
fig <- fig %>% layout( # do.call soon
title = "Top 20 High-Risk Employees, according to RSF's Cumulative Risk in Five Years",
shapes = list(baseline),
xaxis = list(title = "Year", range = c(0, 5), zerolinecolor = 'white'),
yaxis = list(title = "Accummulated Risk over Time", range = c(0, 1.2), zerolinecolor = 'white'),
margin = list(l = 50, t = 50, b = 50, r = 50),
plot_bgcolor = '#e5ecf6',
annotations = list(
list(
xanchor = "left", yanchor = "top", x = -0.5, y = 1, xref = "paper", yref = "paper",
text = "Method", align = "left", showarrow = FALSE
), list(
xanchor = "left", yanchor = "top", x = -0.5, y = 0.8, xref = "paper", yref = "paper",
text = "Based on", align = "left", showarrow = FALSE
)
),
updatemenus = list(
list(
xanchor = "left", yanchor = "top", x = -0.5, y = 0.95, xref = "paper", yref = "paper",
showactive = TRUE,
buttons = list(
list(
label = "Random Survival Forest",
method = "update",
args = list(list(visible = c(rep(TRUE, cutoff), FALSE, TRUE)), rsf_layout, 1:(cutoff+2) - 1)
),
list(
label = "Cox Proportional Hazard",
method = "update",
args = list(list(visible = c(rep(FALSE, cutoff), TRUE, FALSE)), coxph_layout, 1:(cutoff+2) - 1)
)
)
), list(
xanchor = "left", yanchor = "top", x = -0.5, y = 0.75, xref = "paper", yref = "paper",
showactive = TRUE,
buttons = create_buttons(c("EmployeeNumber", "EnvironmentSatisfaction", "JobRole", "MonthlyIncome", "NumCompaniesWorked",
"NotWorkingYears", "OverTime", "StockOptionLevel", "TrainingTimesLastYear", "WorkLifeBalance"))
)
)
)
# Build the HTML widget, only for the case of nbviewer
# Otherwise, simply the line `fig` will suffice
#htmlwidgets::saveWidget(as_widget(fig), "plot.html")
#IRdisplay::display_html("<iframe seamless src='plot.html' width=1000, height=600></iframe>")
fig