-
Notifications
You must be signed in to change notification settings - Fork 0
/
masterclass4.py
212 lines (131 loc) · 6.46 KB
/
masterclass4.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
import pandas as pd
import numpy as np
import sklearn as sk
import seaborn as sns
import matplotlib.pyplot as plt
bank_data = pd.read_csv('', sep=';')
bank_data2= bank_data.copy()
bank_data2.info()
bank_data2['y new'] = bank_data2.y.map(dict(yes=1, no=0));
bank_data2.head()
bank_data2.describe()
#median is greater than mean -> there is a skew in the data
bank_data2.age.median()
plt.style.use('seaborn-whitegrid')
bank_data2.hist(bins=20, figsize=(14,10))
plt.show()
labels = 'Did not open term', 'opened term'
fig, ax = plot.subplots(1,2, figsize=(16,8))
bank_data2['y'].value_counts().plot.pie(explode=[0,0.25], autopct='%1.2f%%', ax=ax[0], shadow=True, labels=labels, fontsize=12, startangle=135)
plt.subtitle('Information on term subscriptions', fontsize=20)
df = bank_data2.groupby(['education', 'y']).size().groupby(level=0).apply(lambda x: x/bank_data2.shape[0]).unstack().plot(kind='bar', ax=ax[1], stacked=True)
ax[1].set(ylabel='Percentage of term openers by level of education')
ax[1].set(xlable='Education level')
ax[1].legend(['Did not open', 'open'])
fig, ax = plt.subplots(1, 2, figsize=(16,8))
plt.subtitle('Information on Term Subscription 2', fontsize=20)
df = bank_data2.groupby(['age', 'y']).size().groupby(level=0, squeeze=True).apply(lambda x: x/bank_rate2.shape[0]).unstack().plot(kind='bar', ax=ax[0], stacked=True)
ax[0].set(ylabel='Percentage of term openers by age')
ax[0].set(xlabel='Age')
ax[0].locator_params(axis='x', nbins=60)
ax[0].legend(['Did not open', 'Open'])
dfl = bank_data2.groupby(['marital', 'y']).size().groupby(level=0).apply(lambda x: x/bank_data2.shape[0]).unstack().plot(kind='bar', ax=ax[1])
ax[1].set(ylabel='Percentage of term openers by marriage status')
ax[1].set(xlabel='Marriage status')
ax[1].legend(['Did not open', 'Open'])
from sklearn.preprocessing import LabelEncoder
encoder = LabelEncoder()
dependent_var = bank_data2['y new']
encoded_df = bank_data2.copy()
encoded_df = encoded_df.drop(['y', 'y new'], axis = 1)
encoded_df = encoded_df.apply(encoder.fit_transform)
encoded_df.head()
encoded_df.describe()
#lets make a correlation matrix
encoded_df = pd.concat([encoded_df, dependent_var], axis=1)
plt.figure(figsize=(12,10))
cor = encoded_df2.corr()
sns.heatmap(cor, annot=True, cmap=plt.cn.Reds)
plt.show()
cor_target = abs(cor["y new"])
relevant_features = cor_target[cor_target > 0.2]
print(relevant_features, '\n')
feature_cors = encoded_df[['duration', 'pdays']].corr()
feature_cors1 = encoded_df[['euribor3m', 'nr.employed']].corr()
feature_cors2 = encoded_df[['job', 'marital']].corr()
feature_cors3 = encoded_df[['marital', 'age']].corr()
print(feature_cors, '\n', feature_cors1, '\n', feature_cors2, '\n', feature_cors3)
#I chose to drop: age, housing, loan, default, day of the week
encoded_df = encoded_df.drop(['age', 'housing', 'loan', 'default', 'day_of_week'], axis = 1)
encoded_df.describe()
print(bank_data2.shape)
train_len = int(.8 * (bank_data2.shape[0]))
train_x, train_y = encoded_df[:train_len], dependent_var[:train_len]
test_x, test_y = encoded_df[train_len:], dependent_var[train_len:]
from sklearn.linear_model import LogistricRegression
log_reg = LogistricRegression(solver='lbfgs', penalty='12', max_iter=10000)
log_reg_trained = log_reg.fit(train_x, train_y)
print(f'training accuracy is: {log_reg.score(train_x, train_y)}')
pred = log_reg.predict(test_x)
print(f'test accuracy is {sk.metrics.accuracy_score(test_y, pred)}')
log_reg.coef
from sklearn.tree import DecisionTreeClassifier
dec_tree = DecisionTreeClassifier()
tree_model = dec_tree.fit(train_x, train_y)
print(f'training accuracy for decision tree is: {dec_tree.score(train_x, train_y)}')
pred1 = dec_tree.predict(test_x)
print(f'training accuracy for decision tree is: {sk.metrics.accuracy_score(test_y, pred1)}')
from aix360.algorithms.lime import LimeTabularExplainer
class_names = [0, 1]
log_lime_explainer = LimeTabularExplainer(train_x.values, class_names=class_names, feature_names=train_x.columns)
print(f'the predicted class is: {log_reg_trained.predict_proba([train_x.values[0]])}')
print(f'the true class is: {train_y.loc[0]}')
idx = 1220
exp_log = log_lime_explainer.explain_instance(train_x.values[idx], log_reg_trained.predict_proba, num_features=6, labels=class_names)
print('Explanation for class %s' %class_names[0])
print('\n'.join(map(str, exp_log.as_list(label=0))))
print('Explanation for class %s' %class_names[1])
print('\n'.join(map(str, exp_log.as_list(label=1))))
log_reg_exp.show_in_notebook()
exp_log.show_in_notebook()
from aix360.metrics import faithfulness_metric, monotonicity_metric
predicted_class = log_reg.predict(test_x.values[0].reshape(1, -1))[0]
le = exp_log.local_exp[predicted_class]
m = exp_log.as_map()
x = test_x.values[0]
coefs = np.zeros(x.shape[0])
for v in le:
coefs[v[0]] = v[1]
base = np.zeros(x.shape[0])
print('Faithfulness:', faithfulness_metric(log_reg, x, coefs, base))
print('Monotonicity:', monotonicity_metric(log_reg, x, coefs, base))
from lale.lib.sklearn import PCA
from lale.lib.sklearn import LogisticRegression
from lale.lib.sklearn import KNeighborsClassifier as Kn
from lale.lib.sklearn import DecisionTreeClassifier as tree
from lale.lib.sklearn import SVC
from lale.lib.sklearn import RandomForestClassifier as rf
from lale.lib.sklearn import XGBClassifier
from lale.lib.sklearn import NoOp, ConcatFeatures
import lale
import lale.helpers
import lale.operators
from lale.lib.lale import Hyperopt
lale.wrap_import_operators()
import sklearn.metrics
sorted(sklearn.metrics.SCORERS.keys())
explainable_pipe = lale.operators.make_union(PCA, NoOp) >> (LogistricRegression | Kn | tree)
explainable_train = explainable_pipe.auto_configure(train_x, train_y, optimizer= Hyperopt, cv= 3, max_evals= 3, scoring= 'accuracy')
explainable_train.visualize()
explainable_train.pretty_print(show_imports= False, ipython_display= True)
print(f'the accuracy of this pipeline is {sklearn.metrics.accuracy_score(test_y, explainable_train.predict(test_x))}')
import aix360
from aix360.algorithms.lime import LimeTabularExplainer
print(f'the predicted class is: {explainable_train.predict([train_x.values[0]])}')
print(f'the true class is {train_y.loc[0]}')
idx = 10
explainable_exp = limeexplainer.explain_instance(train_x.values[idx], explainable_train.predict_proba, num_features= 5, labels= [0, 1])
print('Explanation for class %s' %class_names[0])
print('\n'.join(map(str, explainable_exp.as_list(label=0))))
print('Explanation for class %s' %class_names[1])
print('\n'.join(map(str, explainable_exp.as_list(label=1))))