short_text_classification/classifier_hyperparameter_gridsearch.R at master · jequihua/short_text_classification · GitHub

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
# load packages
library("Matrix")
library("tm")
library("tidyverse")
library("purrrlyr")
library("text2vec")
library("qdap")
library("slam")
library("xgboost")
#library("caret")
library("dummies")

# loud source functions
source("./tools.R")

# load full_table
load("./data/full_corpus.RData")

# shuffle rows
full_table = shuffle_df(full_table)

# class counts imbalanced and a little bit dirty
table(full_table$Tipo)

# clean "Tipo" variable
full_table$Tipo[full_table$Tipo=="au"]="AU"
full_table$Tipo[full_table$Tipo=="ed"]="ED"
full_table$Tipo[full_table$Tipo=="re"]="RE"
full_table$Tipo[full_table$Tipo=="S"]="SA"

# create numerical dependent variable for classification
# as we will use XGBOOST classes must begin at 0
full_table$target = as.numeric(as.factor(full_table$Tipo))-1

# clean "Estado" variable

length(table(full_table$Estado))

full_table$Estado[full_table$Estado=="Veracruz"]="Veracruz de Ignacio de la Llave"
full_table$Estado[full_table$Estado=="Michoacán"]="Michoacán de Ocampo"
full_table$Estado[full_table$Estado=="AGS"]="Aguascalientes"
full_table$Estado[full_table$Estado=="Quintana Roo."]="Quintana Roo"
full_table$Estado[full_table$Estado=="Distrito Federal"]="Ciudad de México"
full_table$Estado[full_table$Estado=="Coahuila"]="Coahuila de Zaragoza"

# make dummy variables from "Estado"
dummies_matrix = dummy(full_table$Estado,sep="_")

### build model training data table

# tokenize
it = itoken(full_table$Destino, tokenizer = word_tokenizer,
            ids = 1:nrow(full_table),
            progressbar = TRUE)

# create vocabulary
vocab = create_vocabulary(it)

vectorizer = vocab_vectorizer(vocab)

# create document-term matrix
dtm = create_dtm(it, vectorizer)

# define tf-idf model
tfidf = TfIdf$new()

### fit the model to data and transform it with the fitted model
dtm_tfidf = fit_transform(dtm, tfidf)

# add estado dummies to matrix
dtm_tfidf = cbind(dtm_tfidf,dummies_matrix)

# split into unlabeled and labeled feature matrix
labeled_matrix = dtm_tfidf[!is.na(full_table$Tipo),]
unlabeled_matrix = dtm_tfidf[is.na(full_table$Tipo),]

total_clean = as.matrix(full_table$Total)
total_clean = total_clean[!is.na(full_table$Tipo)]

labeled_matrix_gasto = cbind(labeled_matrix,total_clean)
labeled_matrix_gasto = labeled_matrix_gasto[total_clean!=0,]

labels = full_table$target[!is.na(full_table$Tipo)]
labels = labels[total_clean!=0]

#sample_idx = sample_class_matrix(labeled_matrix_gasto,labels,100)

#labeled_matrix_gasto_s = labeled_matrix_gasto[sample_idx,]
#labels_s = labels[sample_idx]

# build XGBoost matrix for training
final_data_matrix = xgb.DMatrix(data = labeled_matrix_gasto,
                                 label = labels)

### grid search

# initialize lists
params_list = list()
cv_run_list = list()
eval_metric_list = list()
eval_metric_list_idx = list()
seed_number_list = list()

for (iter in 1:1000) {

  print(iter)

  param = list(objective = "multi:softprob",
                eval_metric = "mlogloss",
                num_class = 32,
                max_depth = sample(3:10, 1),
                eta = runif(1, .01, 0.3),
                gamma = runif(1, 0.0, 0.2),
                subsample = runif(1, .6, .9),
                colsample_bytree = runif(1, .5, .8),
                min_child_weight = sample(1:40, 1),
                max_delta_step = sample(1:10, 1)
  )

  cv.nround = 300
  cv.nfold = 10

  seed.number = sample.int(10000, 1)[[1]]
  set.seed(seed.number)

  mdcv = xgb.cv(data=final_data_matrix ,
                 params = param,
                 nfold=cv.nfold,
                 nrounds=cv.nround,
                 verbose = FALSE,
                 early_stopping_rounds=10,
                 maximize=FALSE,
                 prediction = FALSE)

  min_logloss = min(mdcv$evaluation_log$train_mlogloss_mean)

  min_logloss_index = which.min(mdcv$evaluation_log$train_mlogloss_mean)

  # populate lists
  params_list[[iter]] = param
  cv_run_list[[iter]] = mdcv
  eval_metric_list[[iter]] = min_logloss
  eval_metric_list_idx[[iter]] = min_logloss_index
  seed_number_list[[iter]] = seed.number

}

final_list = list(params_list=params_list,
                  cv_run_list=cv_run_list,
                  eval_metric_list=eval_metric_list,
                  eval_metric_list_idx=eval_metric_list_idx,
                  seed_number_list=seed_number_list)

save(final_list, file="./grid_search_results/grid_search_v1_mlogloss.RData")

#nround = best_logloss_index
#set.seed(best_seednumber)
#md <- xgb.train(data=dtrain, params=best_param, nrounds=nround)

?xgb.train