forked from Guangyi-Zhang/diverse-rule-set
/
models.py
177 lines (144 loc) · 7.02 KB
/
models.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
import numpy as np
import pandas as pd
from functools import partial
import itertools
import Orange
from Orange.data import Domain, DiscreteVariable
from sklearn import tree
from sklearn.metrics import precision_score, recall_score, f1_score, \
roc_auc_score, balanced_accuracy_score, confusion_matrix
roc_auc_score = partial(roc_auc_score, multi_class='ovo', average='macro')
from decisionset import DecisionSet
from itemset import Itemset, Rule, Transaction, prep_db, feat2item
from util import recall, precision, dispersion, printRules, dispersion_, overlap
from util import max_freq_ratio, tree2rules
from ids import IDS, IDS_predict
from pyarc import CBA, TransactionDB
def run_ids(Xtr, Ytr, Xt, Yt, lb, min_freq, lambs, log=None):
ids, nfreq, default = IDS(Xtr, Ytr.values, lambs, freq=min_freq)
for r in ids:
print('class: ', r.class_label, ', cover: {}/{}'.format(len(r.get_correct_cover(Xtr, Ytr)),
len(r.get_cover(Xtr))), end='; ')
r.print_rule()
for r in ids:
r.covered = set(r.get_cover(Xtr))
Y_pred = IDS_predict(ids, Xt, default=default)
if log is None:
from logger import log
[log('ids-lambda', lamb, i) for i, lamb in enumerate(lambs)]
log('ids-k', len(ids))
[log('ids-nconds', r.get_length(), i) for i, r in enumerate(ids)]
log('ids-nfreq', nfreq)
log('ids-freq', min_freq)
log('ids-default', default)
log('ids-auc', roc_auc_score(lb.transform(Yt.values), lb.transform(Y_pred)))
log('ids-bacc', balanced_accuracy_score(Yt, Y_pred))
log('ids-disp', dispersion_(ids, average=True))
log('ids-overlap', overlap(ids))
print(confusion_matrix(Yt, Y_pred))
return Y_pred
def run_cn2(Xtr, Ytr, Xt, Yt, lb, k=None, log=None):
domainx = Domain.from_numpy(Xtr.values)
domainy = Domain.from_numpy(Ytr.values.reshape((-1, 1)))
datax = Orange.data.Table.from_numpy(domainx, Xtr.values)
datay = Orange.data.Table.from_numpy(domainy, Ytr.values.reshape((-1, 1)))
discretizer = Orange.preprocess.DomainDiscretizer()
domainx = discretizer(datax)
domainy = discretizer(datay)
domain = Domain(domainx.attributes, domainy.attributes[0])
data = Orange.data.Table.from_numpy(domain, Xtr.values, Y=Ytr.values)
learner = Orange.classification.CN2UnorderedLearner()
#learner = Orange.classification.rules.CN2Learner()
learner.rule_finder.search_algorithm.beam_width = 10
learner.rule_finder.search_strategy.constrain_continuous = True
learner.rule_finder.general_validator.min_covered_examples = 15
cn2 = learner(data)
if k is not None:
r_def = cn2.rule_list[-1]
cn2.rule_list = cn2.rule_list[:k]
cn2.rule_list.append(r_def)
Y_pred = np.argmax(cn2.predict(Xt.values), axis=1)
ids = np.arange(Xtr.shape[0])
print('default:', cn2.rule_list[-1].prediction)
# Skip the last default rule
for i,r in enumerate(cn2.rule_list[:-1]):
cov = np.array([r.evaluate_instance(x) for x in data])
pred = np.array([r.prediction] * sum(cov))
acc = pred == Ytr.values[cov]
r.covered = set(ids[cov])
print('CN2', '#{}, label:{}, len:{}, cov:{}, acc:{}'.format(i, r.prediction, r.length, sum(cov)/len(ids), sum(acc)/sum(cov)))
if log is None:
from logger import log
log('cn2-k', len(cn2.rule_list[:-1]))
[log('cn2-nconds', r.length, i) for i, r in enumerate(cn2.rule_list[:-1])]
log('cn2-auc', roc_auc_score(lb.transform(Yt.values), lb.transform(Y_pred)))
log('cn2-bacc', balanced_accuracy_score(Yt, Y_pred))
log('cn2-disp', dispersion_(cn2.rule_list[:-1], average=True))
log('cn2-overlap', overlap(cn2.rule_list[:-1]))
print(confusion_matrix(Yt, Y_pred))
def run_cart(Xtr, Ytr, Xt, Yt, lb, k=None, log=None):
clf = tree.DecisionTreeClassifier(criterion='gini', max_depth=None, max_leaf_nodes=k)
clf = clf.fit(Xtr, Ytr)
print('tree depth and #leaf: ', clf.get_depth(), clf.get_n_leaves())
leaves = tree2rules(clf, Xtr.columns)
Y_pred = clf.predict(Xt)
if log is None:
from logger import log
log('CART-k', clf.get_n_leaves())
[log('CART-nconds', len(r.split('AND')), i) for i, r in enumerate(leaves.split('\n'))]
log('CART-depth', clf.get_depth())
log('CART-rules', leaves)
log('CART-auc', roc_auc_score(lb.transform(Yt.values), lb.transform(Y_pred)))
log('CART-bacc', balanced_accuracy_score(Yt, Y_pred))
log('CART-overlap', 0)
print(confusion_matrix(Yt, Y_pred))
def run_cba(Xtr, Ytr, Xt, Yt, lb, support=0.20, confidence=0.5, k=None, log=None):
txns_train = TransactionDB.from_DataFrame(pd.concat([Xtr, Ytr], axis=1))
txns_test = TransactionDB.from_DataFrame(pd.concat([Xt, Yt], axis=1))
cba = CBA(support=support, confidence=confidence, algorithm="m1")
cba.fit(txns_train)
if k is not None:
cba.clf.rules = cba.clf.rules[:k]
Y_pred = [int(i) for i in cba.predict(txns_test)]
for r in cba.clf.rules:
r.covered = set([i for i, rd in enumerate(txns_train) if r.antecedent <= rd])
if log is None:
from logger import log
log('cba-k', len(cba.clf.rules))
log('cba-rules', str(cba.clf.rules))
[log('cba-nconds', len(r), i) for i, r in enumerate(cba.clf.rules)]
log('cba-auc', roc_auc_score(lb.transform(Yt.values), lb.transform(Y_pred)))
log('cba-bacc', balanced_accuracy_score(Yt, Y_pred))
log('cba-disp', dispersion_(cba.clf.rules, average=True))
log('cba-overlap', overlap(cba.clf.rules))
print(confusion_matrix(Yt, Y_pred))
def run_ours(Xtr, Ytr, Xt, Yt, lb, nsample, lambda_mode, q, sample_mode, k=None, rerun=True,
eps=0.01, min_recall_per_class=0.8,
log=None):
#name = 'ours' if k is None else 'oursk'
name = 'ours{}'.format(int(rerun))
k = k if k is not None else 100
dec = DecisionSet(eps)
dec.train(Xtr, Ytr, max_k=k, nsamp=nsample, lamb=lambda_mode, q=q, mode=sample_mode, rerun=rerun, min_recall_per_class=min_recall_per_class)
print('default:', dec.default)
Xt_ = [Transaction(feat2item(t)) for t in Xt.values]
Y_pred = dec.predict_all(Xt_)
if log is None:
from logger import log
log('{}-default'.format(name), dec.default)
log('{}-k'.format(name), len(dec.rules))
log('{}-maxk'.format(name), k)
[log('{}-nconds'.format(name), len(r), i) for i, r in enumerate(dec.rules)]
log('{}-q'.format(name), q)
log('{}-nsample'.format(name), nsample)
log('{}-lamb'.format(name), lambda_mode)
log('{}-seq'.format(name), dec.seq)
log('{}-auc'.format(name), roc_auc_score(lb.transform(Yt.values), lb.transform(Y_pred)))
log('{}-bacc'.format(name), balanced_accuracy_score(Yt, Y_pred))
log('{}-disp'.format(name), dispersion(dec.rules, average=True))
log('{}-overlap'.format(name), overlap(dec.rules))
log('{}-mode'.format(name), sample_mode)
[log('{}-precisions-tr'.format(name), v, l) for l, v in precision(dec).items()]
[log('{}-recall-tr'.format(name), v, l) for l, v in recall(dec.rules).items()]
print(confusion_matrix(Yt, Y_pred))
return Y_pred