-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtest_multi_cost_sensitive_alarm2.py
251 lines (202 loc) · 14 KB
/
test_multi_cost_sensitive_alarm2.py
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
import EncoderFactory
from DatasetManager import DatasetManager
import pandas as pd
import numpy as np
from sklearn.metrics import roc_auc_score, precision_recall_fscore_support, confusion_matrix
from sklearn.pipeline import FeatureUnion
import time
import os
import sys
import csv
from sys import argv
import pickle
from conf_constant_costfunctions import get_constant_costfunctions
import xgboost as xgb
def calculate_cost(x, costs):
return costs[int(x['prediction']), int(x['actual'])](x)
def calculate_cost_baseline(x, costs):
return costs[0, int(x['actual'])](x)
def cost_sensitive_eval(y_predicted, y_true):
labels = y_true.get_label()
# return a pair metric_name, result
#false negative: abs(c_in_alarm2 - c_in_alarm1)
# abs(1.2*2-2)
fn = sum((labels >= 0.5) & (y_predicted < 0.5)) * cost_false_negative
#false positive abs((c_in_alarm2 + c_com_alarm2) - (c_in_alarm1 + c_com_alarm2))
# 1.2 + 5 - 1 + 10
fp = sum((labels <= 0.5) & (y_predicted > 0.5)) * cost_false_positive
return 'cost', (fn + fp)
dataset_name = argv[1]
predictions_dir = argv[2]
conf_threshold_dir = argv[3]
train_data_dir = argv[4]
optimal_params_dir = argv[5]
results_dir = argv[6]
trial_nr = 1
method = "cost_sensitive_multi"
# create results directory
if not os.path.exists(os.path.join(results_dir)):
os.makedirs(os.path.join(results_dir))
# load predictions
dt_preds = pd.read_csv(os.path.join(predictions_dir, "preds_%s.csv" % dataset_name), sep=";")
# write results to file
out_filename = os.path.join(results_dir, "results_%s_%s.csv" % (dataset_name, method))
# set nonomonotic constants
aConst, bConst, cConst, dConst, eConst, fConst = get_constant_costfunctions(dataset_name)
with open(out_filename, 'w') as fout:
writer = csv.writer(fout, delimiter=';', quotechar='', quoting=csv.QUOTE_NONE)
writer.writerow(["dataset", "method", "metric", "value", "c_miss", "c_action", "c_postpone", "c_com", "early_type",
"threshold"])
cost_weights = [(10, 1), (10, 2)]
c_com_weights = [1, 2, 3, 4, 5, 10, 20, 30, 40]
c_postpone_weight = 0
alarm_type = ["alarm1","alarm2"]
for c_miss_weight, c_action_weight in cost_weights:
for c_com_weight in c_com_weights:
for early_type in ["const"]:
for v_alarm_type in alarm_type:
# Alarm2 instead of 1
cost_false_positive = float(c_action_weight) * 0.2
# Alarm 1 instead of 2
cost_false_negative = (float(c_action_weight) + float(c_com_weight)) - (
(1.2 * float(c_action_weight)) + (float(c_com_weight) * 0.5))
train_data_file = os.path.join(train_data_dir, "filtered_events_%s_%s_%s_%s_%s_%s_%s.csv" % (
dataset_name,v_alarm_type, c_miss_weight, c_action_weight, c_postpone_weight, c_com_weight, early_type))
optimal_params_filename = os.path.join(optimal_params_dir, "xgboost_params_%s_%s_%s_%s_%s_%s_%s.pickle" % (
dataset_name,v_alarm_type, c_miss_weight, c_action_weight, c_postpone_weight, c_com_weight, early_type))
print(optimal_params_filename)
if os.path.exists(optimal_params_filename):
###############Initiate XGB for alarm choice#######
# read the data
dataset_manager = DatasetManager(dataset_name)
data = dataset_manager.read_dataset_file(train_data_file)
min_prefix_length = 1
max_prefix_length = int(np.ceil(data.groupby(dataset_manager.case_id_col).size().quantile(0.9)))
cls_encoder_args = {'case_id_col': dataset_manager.case_id_col,
'static_cat_cols': dataset_manager.static_cat_cols,
'static_num_cols': dataset_manager.static_num_cols,
'dynamic_cat_cols': dataset_manager.dynamic_cat_cols,
'dynamic_num_cols': dataset_manager.dynamic_num_cols,
'fillna': True}
train_ratio = 0.85
# generate data where each prefix is a separate instance
train, val = dataset_manager.split_val(data, train_ratio)
dt_train = dataset_manager.generate_prefix_data(train, min_prefix_length, max_prefix_length)
dt_val = dataset_manager.generate_prefix_data(val, min_prefix_length, max_prefix_length)
# train the model with pre-tuned parameters
with open(optimal_params_filename, "rb") as fin:
best_params = pickle.load(fin)
# encode all prefixes
feature_combiner = FeatureUnion(
[(method, EncoderFactory.get_encoder(method, **cls_encoder_args)) for method in
["static", "agg"]])
X_train = feature_combiner.fit_transform(dt_train)
y_train = np.array(dataset_manager.get_label_numeric(dt_train))
X_val = feature_combiner.fit_transform(dt_val)
y_val = np.array(dataset_manager.get_label_numeric(dt_val))
print(best_params)
cls = xgb.XGBClassifier(objective='binary:logistic',
n_estimators=best_params['n_estimators'],
learning_rate=best_params['learning_rate'],
subsample=best_params['subsample'],
max_depth=int(best_params['max_depth']),
colsample_bytree=best_params['colsample_bytree'],
min_child_weight=int(best_params['min_child_weight']),
seed=22,
n_jobs=-1)
eval_set = [(X_val, y_val)]
cls.fit(X_train, y_train, eval_set=eval_set, early_stopping_rounds=10,
eval_metric=cost_sensitive_eval)
print("Done training classifier")
###############End XGB for alarm choice#######
c_miss = c_miss_weight / (c_miss_weight + c_action_weight + c_com_weight)
c_action = c_action_weight / (c_miss_weight + c_action_weight + c_com_weight)
c_com = c_com_weight / (c_miss_weight + c_action_weight + c_com_weight)
c_action2 = c_action * 1.2
c_com2 = c_com * 0.5
costs = np.matrix([[lambda x: 0,
lambda x: c_miss],
[lambda x: c_action2 + c_com2,
lambda x: c_action2
],
[lambda x: c_action + c_com,
lambda x: c_action
]])
# load the optimal confidence threshold
conf_file = os.path.join(conf_threshold_dir, "optimal_confs_%s_%s_%s_%s_%s_%s.pickle" % (
dataset_name, c_miss_weight, c_action_weight, c_postpone_weight, c_com_weight, early_type))
with open(conf_file, "rb") as fin:
conf_threshold = pickle.load(fin)['conf_threshold']
# trigger alarms according to conf_threshold
dt_final = pd.DataFrame()
unprocessed_case_ids = set(dt_preds.case_id.unique())
for nr_events in range(1, dt_preds.prefix_nr.max() + 1):
tmp = dt_preds[(dt_preds.case_id.isin(unprocessed_case_ids)) & (dt_preds.prefix_nr == nr_events)]
tmp = tmp[tmp.predicted_proba >= conf_threshold]
tmp["prediction"] = 1
dt_final = pd.concat([dt_final, tmp], axis=0)
unprocessed_case_ids = unprocessed_case_ids.difference(tmp.case_id)
# read the data
dataset_manager_multi = DatasetManager(dataset_name)
data_multi = dataset_manager_multi.read_dataset()
dt_multi = pd.DataFrame()
dt_multi = data_multi[(data_multi["Case ID"].isin(dt_final["case_id"])) & (data_multi["event_nr"].isin(dt_final["prefix_nr"]))]
X_test = feature_combiner.fit_transform(dt_multi)
print("Feature Combiner sucessful")
dt_final["prediction"] = cls.predict(X_test) + 1
print(dt_final["prediction"].value_counts())
tmp = dt_preds[(dt_preds.case_id.isin(unprocessed_case_ids)) & (dt_preds.prefix_nr == 1)]
tmp["prediction"] = 0
dt_final = pd.concat([dt_final, tmp], axis=0)
case_lengths = dt_preds.groupby("case_id").prefix_nr.max().reset_index()
case_lengths.columns = ["case_id", "case_length"]
dt_final = dt_final.merge(case_lengths)
# calculate precision, recall etc.
prec, rec, fscore, _ = precision_recall_fscore_support(dt_final.actual, dt_final.prediction,
pos_label=1, average="macro")
#tn, fp, fn, tp = confusion_matrix(dt_stasts.actual, dt_final.prediction).ravel()
# calculate earliness based on the "true alarms" only
tmp = dt_final[(dt_final.prediction == 1) & (dt_final.actual == 1)]
earliness = (1 - ((tmp.prefix_nr - 1) / tmp.case_length))
tmp = dt_final[(dt_final.prediction == 1)]
earliness_alarms = (1 - ((tmp.prefix_nr - 1) / tmp.case_length))
writer.writerow([dataset_name, method, "prec", prec, c_miss_weight, c_action_weight, c_postpone_weight,
c_com_weight, early_type, conf_threshold])
writer.writerow(
[dataset_name, method, "rec", rec, c_miss_weight, c_action_weight, c_postpone_weight, c_com_weight,
early_type, conf_threshold])
writer.writerow(
[dataset_name, method, "fscore", fscore, c_miss_weight, c_action_weight, c_postpone_weight,
c_com_weight, early_type, conf_threshold])
# writer.writerow(
# [dataset_name, method, "tn", tn, c_miss_weight, c_action_weight, c_postpone_weight, c_com_weight,
# early_type, conf_threshold])
# writer.writerow(
# [dataset_name, method, "fp", fp, c_miss_weight, c_action_weight, c_postpone_weight, c_com_weight,
# early_type, conf_threshold])
# writer.writerow(
# [dataset_name, method, "fn", fn, c_miss_weight, c_action_weight, c_postpone_weight, c_com_weight,
# early_type, conf_threshold])
# writer.writerow(
# [dataset_name, method, "tp", tp, c_miss_weight, c_action_weight, c_postpone_weight, c_com_weight,
# early_type, conf_threshold])
writer.writerow(
[dataset_name, method, "earliness_mean", earliness.mean(), c_miss_weight, c_action_weight,
c_postpone_weight, c_com_weight, early_type, conf_threshold])
writer.writerow([dataset_name, method, "earliness_std", earliness.std(), c_miss_weight, c_action_weight,
c_postpone_weight, c_com_weight, early_type, conf_threshold])
writer.writerow([dataset_name, method, "earliness_alarms_mean", earliness_alarms.mean(), c_miss_weight,
c_action_weight, c_postpone_weight, c_com_weight, early_type, conf_threshold])
writer.writerow([dataset_name, method, "earliness_alarms_std", earliness_alarms.std(), c_miss_weight,
c_action_weight, c_postpone_weight, c_com_weight, early_type, conf_threshold])
cost = dt_final.apply(calculate_cost, costs=costs, axis=1).sum()
writer.writerow([dataset_name, method, "cost", cost, c_miss_weight, c_action_weight, c_postpone_weight,
c_com_weight, early_type, conf_threshold])
writer.writerow([dataset_name, method, "cost_avg", cost / len(dt_final), c_miss_weight, c_action_weight,
c_postpone_weight, c_com_weight, early_type, conf_threshold])
cost_baseline = dt_final.apply(calculate_cost_baseline, costs=costs, axis=1).sum()
writer.writerow([dataset_name, method, "cost_baseline", cost_baseline, c_miss_weight, c_action_weight,
c_postpone_weight, c_com_weight, early_type, conf_threshold])
writer.writerow(
[dataset_name, method, "cost_avg_baseline", cost_baseline / len(dt_final), c_miss_weight,
c_action_weight, c_postpone_weight, c_com_weight, early_type, conf_threshold])