prescriptiveProcessMonitoring/optimize_multiple_thresholds_compensation.py at master · samadeusfp/prescriptiveProcessMonitoring · 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
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
import EncoderFactory
from DatasetManager import DatasetManager

import pandas as pd
import numpy as np

from sklearn.metrics import roc_auc_score
from sklearn.pipeline import FeatureUnion

import time
import os
import sys
from sys import argv
import pickle
import csv

from hyperopt import Trials, STATUS_OK, tpe, fmin, hp
import hyperopt
from multiprocessing import Process as Process


def calculate_expected_costs(x, costs, alarm):
    return ((x.predicted_proba) * costs[alarm,1](x)) + ((1 - x.predicted_proba) * costs[alarm,0](x))

def get_max_nonmonotonic(dataSetName):
    if str.startswith(dataSetName,"traffic_fines"):
        return 2
    elif str.startswith(dataSetName,"bpic2017"):
        return 12

def calculate_cost(x, costs):
    actual = int(x['actual'])
    if int(x['prediction']) == 1:
        if x["alarm1"] == 1 and x["alarm2"] == 1:
            if calculate_expected_costs(x,costs,1) < calculate_expected_costs(x,costs,1):
                optimalCosts = costs[1, actual](x)
            else:
                optimalCosts = costs[2, actual](x)
        elif x["alarm1"] == 1:
            optimalCosts = costs[1, actual](x)
        elif x["alarm2"] == 1:
            optimalCosts = costs[2, actual](x)
    else:
        optimalCosts = costs[0, actual](x)
    return optimalCosts

def get_min_nonmonotonic(dataSetName):
    if str.startswith(dataSetName,"traffic_fines"):
        return 3
    elif str.startswith(dataSetName,"bpic2017"):
        return 18

def evaluate_model_cost(args):
    conf_threshold_1 = args['conf_threshold_1']
    conf_threshold_2 = args['conf_threshold_2']
    c_action = args['c_action']
    c_miss = args['c_miss']
    c_com = args['c_com']


    if early_type == "linear":
        costs = np.matrix([[lambda x: 0,
                            lambda x: c_miss],
                           [lambda x: c_action * (x['prefix_nr'] - 1) / x['case_length'] + c_com,
                            lambda x: c_action * (x['prefix_nr'] - 1) / x['case_length'] + (
                                    x['prefix_nr'] - 1) / x['case_length'] * c_miss
                            ],
                           [lambda x: c_action * 3 * (x['prefix_nr'] - 1) / x['case_length'] + c_com / 2,
                            lambda x: c_action * 3 * (x['prefix_nr'] - 1) / x['case_length'] + (
                                    x['prefix_nr'] - 1) / x['case_length'] * c_miss
                            ]])
    elif early_type == "nonmonotonic":
        costs = np.matrix([[lambda x: 0,
                            lambda x: c_miss],
                           [lambda x: c_action * (x['case_length'] - min(x['prefix_nr'], nonmonotonic_threshold)/ x['case_length']) + (c_com * (x['case_length'] - x['prefix_nr']/ x['case_length'])),
                            lambda x: c_action * (x['case_length'] - min(x['prefix_nr'], nonmonotonic_threshold)/ x['case_length']) + max((
                                    x['prefix_nr'] - 1) / x['case_length'],max_nonmonotonic_threshold) * c_miss
                            ],
                           [lambda x: c_action * (x['case_length'] - min(x['prefix_nr'], nonmonotonic_threshold)/ x['case_length']) * 3 + (c_com * (x['case_length'] - x['prefix_nr']/ x['case_length'])) / 2,  # 0:2
                            lambda x: c_action * (x['case_length'] - min(x['prefix_nr'], nonmonotonic_threshold)/ x['case_length']) * 3 + max((
                                    x['prefix_nr'] - 1) / x['case_length'],max_nonmonotonic_threshold) * c_miss
                            ]
                           ])
    else:
        costs = np.matrix([[lambda x: 0,
                            lambda x: c_miss],
                           [lambda x: c_action + c_com,  # 0:1
                            lambda x: c_action + (x['prefix_nr'] - 1) / x['case_length'] * c_miss
                            ],
                           [lambda x: c_action * 3 + c_com / 2,  # 0:2
                            lambda x: c_action * 3 + (x['prefix_nr'] - 1) / x['case_length'] * c_miss
                            ]])

    # trigger alarms according to conf_threshold
    dt_final = pd.DataFrame()
    unprocessed_case_ids_alarm1 = set(dt_preds.case_id.unique())
    unprocessed_case_ids_alarm2 = set(dt_preds.case_id.unique())
    unprocessed_case_ids = set(dt_preds.case_id.unique())
    #alarm2
    for nr_events in range(1, dt_preds.prefix_nr.max() + 1):
        tmp = dt_preds[(dt_preds.case_id.isin(unprocessed_case_ids_alarm2)) & (dt_preds.prefix_nr == nr_events)]
        tmp = tmp[tmp.predicted_proba >= conf_threshold_2]
        tmp["prediction"] = 1
        tmp["alarm2"] = 1
        dt_final = pd.concat([dt_final, tmp], axis=0)
        unprocessed_case_ids_alarm2 = unprocessed_case_ids_alarm2.difference(tmp.case_id)
        unprocessed_case_ids = unprocessed_case_ids.difference(tmp.case_id)
    #alarm1
    for nr_events in range(1, dt_preds.prefix_nr.max() + 1):
        tmp = dt_preds[(dt_preds.case_id.isin(unprocessed_case_ids_alarm1)) & (dt_preds.prefix_nr == nr_events)]
        tmp = tmp[tmp.predicted_proba >= conf_threshold_1]
        tmp["prediction"] = 1
        tmp["alarm1"] = 1
        dt_final = pd.concat([dt_final, tmp], axis=0)
        unprocessed_case_ids_alarm1 = unprocessed_case_ids_alarm1.difference(tmp.case_id)
        unprocessed_case_ids = unprocessed_case_ids.difference(tmp.case_id)
    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)

    cost = dt_final.apply(calculate_cost, costs=costs, axis=1).sum()

    return {'loss': cost, 'status': STATUS_OK, 'model': dt_final}

def run_experiment(c_miss_weight,c_action_weight,c_com_weight,early_type):
    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_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)


    space = {'conf_threshold_1': hp.uniform("conf_threshold_1", 0, 1),
             'conf_threshold_2': hp.uniform("conf_threshold_2", 0, 1),
             'c_action': c_action,
             'c_miss': c_miss,
             'c_com': c_com}
    trials = Trials()
    best = fmin(evaluate_model_cost, space, algo=tpe.suggest, max_evals=100, trials=trials)

    best_params = hyperopt.space_eval(space, best)

    outfile = os.path.join(params_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))
    # write to file
    with open(outfile, "wb") as fout:
        print(outfile)
        print(repr(best_params))
        pickle.dump(best_params, fout)


print('Preparing data...')
start = time.time()

dataset_name = argv[1]
preds_dir = argv[2]
params_dir = argv[3]

# create output directory
if not os.path.exists(os.path.join(params_dir)):
    os.makedirs(os.path.join(params_dir))

# read the data
dataset_manager = DatasetManager(dataset_name)

# prepare the dataset
dt_preds = pd.read_csv(os.path.join(preds_dir, "preds_val_%s.csv" % dataset_name), sep=";")

#set nonomonotic-threshold
nonmonotonic_threshold = get_min_nonmonotonic(dataset_name)
max_nonmonotonic_threshold = get_max_nonmonotonic(dataset_name)


print('Optimizing parameters...')
cost_weights = [(1, 1), (2, 1), (3, 1), (5, 1), (10, 1), (20, 1), (40, 1)]
c_com_weights = [1 / 40.0, 1 / 20.0, 1 / 10.0, 1 / 5.0, 1 / 3.0, 1 / 2.0, 1, 2, 3, 5, 10, 20, 40, 0]
#cost_weights = [(10, 1)]
#c_com_weights = [2]
c_postpone_weight = 0
processes = []
for c_miss_weight, c_action_weight in cost_weights:
    for c_com_weight in c_com_weights:
        for early_type in ["const", "linear","nonmonotonic"]:
             p = Process(target=run_experiment,args=(c_miss_weight, c_action_weight, c_com_weight, early_type))
             p.start()
             processes.append(p)
for p in processes:
    p.join()