-
Notifications
You must be signed in to change notification settings - Fork 2k
/
pyunit_all_confusion_matrix_funcs.py
140 lines (118 loc) · 6.2 KB
/
pyunit_all_confusion_matrix_funcs.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
from __future__ import print_function
from builtins import range
import sys
sys.path.insert(1,"../../")
import h2o
from tests import pyunit_utils
import random
from h2o.estimators.gbm import H2OGradientBoostingEstimator
def all_confusion_matrix_funcs():
metrics = ["recall", "specificity", "min_per_class_accuracy", "absolute_mcc", "precision", "accuracy", "f0point5", "f2", "f1", "mean_per_class_accuracy"]
train = [True, False]
valid = [True, False]
print("PARSING TRAINING DATA")
air_train = h2o.import_file(path=pyunit_utils.locate("smalldata/airlines/AirlinesTrain.csv.zip"))
air_train["IsDepDelayed"] = air_train["IsDepDelayed"].asfactor()
print("PARSING TESTING DATA")
air_test = h2o.import_file(path=pyunit_utils.locate("smalldata/airlines/AirlinesTest.csv.zip"))
air_test["IsDepDelayed"] = air_test["IsDepDelayed"].asfactor()
print()
print("RUNNING FIRST GBM: ")
print()
gbm_bin = H2OGradientBoostingEstimator(distribution="bernoulli")
gbm_bin.train(x=["Origin", "Dest", "Distance", "UniqueCarrier", "fMonth", "fDayofMonth","fDayOfWeek"],
y="IsDepDelayed", training_frame=air_train, validation_frame=air_test)
print()
print("RUNNING SECOND GBM: ")
print()
air_train["fDayOfWeek"] = air_train["fDayOfWeek"].asfactor()
air_test["fDayOfWeek"] = air_test["fDayOfWeek"].asfactor()
gbm_mult = H2OGradientBoostingEstimator( distribution="multinomial")
gbm_mult.train(x=["Origin", "Dest", "Distance", "UniqueCarrier", "IsDepDelayed", "fDayofMonth", "fMonth"],
y="fDayOfWeek", training_frame=air_train, validation_frame=air_test)
def dim_check(cm, m, t, v):
assert len(cm) == 2 and len(cm[0]) == 2 and len(cm[1]) == 2, "incorrect confusion matrix dimensions " \
"for metric/thresh: {0}, train: {1}, valid: " \
"{2}".format(m, t, v)
def type_check(cm, m, t, v):
assert isinstance(cm[0][0], (int, float)) and isinstance(cm[0][1], (int, float)) and \
isinstance(cm[1][0], (int, float)) and isinstance(cm[0][0], (int, float)), \
"confusion matrix entries should be integers or floats but got {0}, {1}, {2}, {3}. metric/thresh: {4}, " \
"train: {5}, valid: {6}".format(type(cm[0][0]), type(cm[0][1]), type(cm[1][0]), type(cm[1][1]), m,
t, v)
def count_check(cm, m, t, v):
if v:
assert cm[0][0] + cm[0][1] + cm[1][0] + cm[1][1] == air_test.nrow, \
"incorrect confusion matrix elements: {0}, {1}, {2}, {3}. Should sum " \
"to {4}. metric/thresh: {5}, train: {6}, valid: {7}".format(cm[0][0], cm[0][1], cm[1][0], cm[1][1],
air_test.nrow, m, t, v)
else:
assert cm[0][0] + cm[0][1] + cm[1][0] + cm[1][1] == air_train.nrow, \
"incorrect confusion matrix elements: {0}, {1}, {2}, {3}. Should sum " \
"to {4}. metric/thresh: {5}, train: {6}, valid: {7}".format(cm[0][0], cm[0][1], cm[1][0], cm[1][1],
air_train.nrow, m, t, v)
# H2OBinomialModel.confusion_matrix()
for m in metrics:
for t in train:
for v in valid:
if t and v: continue
cm = gbm_bin.confusion_matrix(metrics=m, train=t, valid=v)
if cm:
cm = cm.to_list()
dim_check(cm, m, t, v)
type_check(cm, m, t, v)
count_check(cm, m, t, v)
# H2OBinomialModel.confusion_matrix()
for x in range(10):
for t in train:
for v in valid:
if t and v: continue
thresholds = [gbm_bin.find_threshold_by_max_metric(m,t,v) for m in
random.sample(metrics,random.randint(1,len(metrics)))]
cms = gbm_bin.confusion_matrix(thresholds=thresholds, train=t, valid=v)
if not isinstance(cms, list): cms = [cms]
for idx, cm in enumerate(cms):
cm = cm.to_list()
dim_check(cm, thresholds[idx], t, v)
type_check(cm, thresholds[idx], t, v)
count_check(cm, thresholds[idx], t, v)
# H2OMultinomialModel.confusion_matrix()
cm = gbm_mult.confusion_matrix(data=air_test)
cm_count = 0
for r in range(7):
for c in range(7):
cm_count += cm.cell_values[r][c]
assert cm_count == air_test.nrow, "incorrect confusion matrix elements. Should sum to {0}, but got {1}".\
format(air_test.nrow, cm_count)
# H2OBinomialModelMetrics.confusion_matrix()
bin_perf = gbm_bin.model_performance(valid=True)
for metric in metrics:
cm = bin_perf.confusion_matrix(metrics=metric).to_list()
dim_check(cm, metric, False, True)
type_check(cm, metric, False, True)
count_check(cm, metric, False, True)
# H2OBinomialModelMetrics.confusion_matrix()
bin_perf = gbm_bin.model_performance(train=True)
for x in range(10):
thresholds = [gbm_bin.find_threshold_by_max_metric(m,t,v) for m in
random.sample(metrics,random.randint(1,len(metrics)))]
cms = bin_perf.confusion_matrix(thresholds=thresholds)
if not isinstance(cms, list): cms = [cms]
for idx, cm in enumerate(cms):
cm = cm.to_list()
dim_check(cm, thresholds[idx], True, False)
type_check(cm, thresholds[idx], True, False)
count_check(cm, thresholds[idx], True, False)
# H2OMultinomialModelMetrics.confusion_matrix()
mult_perf = gbm_mult.model_performance(valid=True)
cm = mult_perf.confusion_matrix()
cm_count = 0
for r in range(7):
for c in range(7):
cm_count += cm.cell_values[r][c]
assert cm_count == air_test.nrow, "incorrect confusion matrix elements. Should sum to {0}, but got {1}". \
format(air_test.nrow, cm_count)
if __name__ == "__main__":
pyunit_utils.standalone_test(all_confusion_matrix_funcs)
else:
all_confusion_matrix_funcs()