/
train_model.py
200 lines (171 loc) · 8.29 KB
/
train_model.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
# Models
from sklearn.model_selection import train_test_split, GridSearchCV
from sklearn.feature_selection import SelectKBest, chi2
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline, make_pipeline
# Sampling
from imblearn.under_sampling import RandomUnderSampler
from imblearn.over_sampling import RandomOverSampler
# Plotting
import matplotlib.pyplot as plt
from sklearn.metrics import f1_score, precision_score, recall_score, confusion_matrix, accuracy_score, plot_precision_recall_curve, plot_roc_curve, plot_confusion_matrix
# IO
import numpy as np
import pandas as pd
import pickle
pd.options.display.float_format = "{:,.2f}".format
# Read dataset into memory
data = pd.read_csv("./JSVulnerabilityDataSet-1.0.csv")
parametersIndex = data.columns.get_loc("CC")
resultIndex = data.columns.get_loc("Vuln")
X = data.iloc[:, parametersIndex:resultIndex]
Y = data.iloc[:, resultIndex]
# X = SelectKBest(chi2, k=20).fit_transform(X, Y)
# print(X_new)
# Prepare plots
fig_confusion, subs_confusion = plt.subplots(3,3)
fig_metrics, subs_metrics = plt.subplots(3,3)
subs_metrics[0][0].set_title('precision vs recall')
subs_metrics[0][1].set_title('ROC')
subs_metrics[0][2].set_title('Accuracy')
subs_confusion[0][0].set_title('no sampling')
subs_confusion[0][1].set_title('over sampling')
subs_confusion[0][2].set_title('under sampling')
for idx, row in enumerate(subs_metrics):
row[0].set_yticklabels([])
row[0].set_xticklabels([])
row[0].get_xaxis().set_visible(False)
row[1].set_yticklabels([])
row[1].set_xticklabels([])
row[1].get_yaxis().set_visible(False)
row[1].get_xaxis().set_visible(False)
row[2].set_yticklabels([])
row[2].set_xticklabels([])
row[2].get_yaxis().set_visible(False)
row[2].get_xaxis().set_visible(False)
for idx, row in enumerate(subs_confusion):
row[0].set_yticklabels([])
row[0].set_xticklabels([])
row[0].get_xaxis().set_visible(False)
row[1].set_yticklabels([])
row[1].set_xticklabels([])
row[1].get_yaxis().set_visible(False)
row[1].get_xaxis().set_visible(False)
row[2].set_yticklabels([])
row[2].set_xticklabels([])
row[2].get_yaxis().set_visible(False)
row[2].get_xaxis().set_visible(False)
row = 0
def train_model(model_type, X_train, Y_train, sampling, grid):
if sampling != "no":
sampler = RandomOverSampler() if sampling == "over" else RandomUnderSampler()
X_train, Y_train = sampler.fit_resample(X_train, Y_train)
trained_model = GridSearchCV(make_pipeline(StandardScaler(), model_type), grid, scoring="f1")
trained_model.fit(X_train, Y_train)
print("Done training model, the following hyper parameters were used:")
print(trained_model.best_params_)
print("\n")
return trained_model
def plot_metric(title, metric_plotter, trained_model, X_test, Y_test, row, col):
plot_configuration = metric_plotter(trained_model, X_test, Y_test)
plot_configuration.plot(ax=subs_metrics[row][col], name=title)
def get_metrics(model, X_test, Y_test):
threshs = []
prec = []
rec = []
f1s = []
for threshold in np.arange(0, 1, 0.1):
prediction = (model.predict_proba(X_test)[:, 1] > threshold).astype('float')
threshs.append(threshold)
prec.append(precision_score(Y_test, prediction, zero_division=0))
rec.append(recall_score(Y_test, prediction))
f1s.append(f1_score(Y_test, prediction))
df = pd.DataFrame([prec, rec, f1s], index=["Precision", "Recall", "F1"])
df.columns = threshs
return df
models = {
"Logistic Regression": {
"model": LogisticRegression(solver='liblinear', max_iter=10000),
"grid": {
'logisticregression__penalty': ['l1','l2'],
'logisticregression__C': [0.001,0.01,0.1,1,10,100,1000],
}
},
"KNN": {
"model": KNeighborsClassifier(),
"grid": {"kneighborsclassifier__n_neighbors": range(1, 10)}
},
"Decision Tree": {
"model": DecisionTreeClassifier(),
"grid": {
'decisiontreeclassifier__max_leaf_nodes': list(range(2, 100)),
'decisiontreeclassifier__min_samples_split': [2, 3, 4]
}
}
}
for model_name, params in models.items():
# Split the data for training and validation
X_train, X_test, Y_train, Y_test = train_test_split(X, Y)
# Train the model with different sampling strategies
print("Training: " + model_name)
print("will preform grid-search for parameters, optimising for F1 score")
print("====================\n\n")
model_over_sampling = train_model(params["model"], X_train, Y_train, "over", params["grid"])
print(get_metrics(model_over_sampling, X_test, Y_test))
get_metrics(model_over_sampling, X_test, Y_test).to_html(open("./tables/" + model_name + "-over.html", "w"))
print("====================")
model_under_sampling = train_model(params["model"], X_train, Y_train, "under", params["grid"])
print(get_metrics(model_under_sampling, X_test, Y_test))
get_metrics(model_over_sampling, X_test, Y_test).to_html(open("./tables/" + model_name + "-under.html", "w"))
print("====================")
model_no_sampling = train_model(params["model"], X_train, Y_train, "no", params["grid"])
print(get_metrics(model_no_sampling, X_test, Y_test))
get_metrics(model_over_sampling, X_test, Y_test).to_html(open("./tables/" + model_name + "-no.html", "w"))
print("====================")
# Report confusion matrices
confusion_plot_no = plot_confusion_matrix(model_no_sampling, X_test, Y_test)
confusion_plot_no.plot(ax=subs_confusion[row][0])
confusion_plot_over = plot_confusion_matrix(model_over_sampling, X_test, Y_test)
confusion_plot_over.plot(ax=subs_confusion[row][1])
confusion_plot_under = plot_confusion_matrix(model_under_sampling, X_test, Y_test)
confusion_plot_under.plot(ax=subs_confusion[row][2])
# Report precision recall curve
plot_metric('no sampling', plot_precision_recall_curve, model_no_sampling, X_test, Y_test, row, 0)
plot_metric('over sampling', plot_precision_recall_curve, model_over_sampling, X_test, Y_test, row, 0)
plot_metric('under sampling', plot_precision_recall_curve, model_under_sampling, X_test, Y_test, row, 0)
# Report ROC curve
plot_metric('no sampling', plot_roc_curve, model_no_sampling, X_test, Y_test, row, 1)
plot_metric('over sampling', plot_roc_curve, model_over_sampling, X_test, Y_test, row, 1)
plot_metric('under sampling', plot_roc_curve, model_under_sampling, X_test, Y_test, row, 1)
# Report accuracy
prediction_no_sampling = model_no_sampling.predict(X_test)
prediction_over_sampling = model_over_sampling.predict(X_test)
prediction_under_sampling = model_under_sampling.predict(X_test)
accuracy_no_sampling = accuracy_score(Y_test, prediction_no_sampling)
accuracy_over_sample = accuracy_score(Y_test, prediction_over_sampling)
accuracy_under_sample = accuracy_score(Y_test, prediction_under_sampling)
subs_metrics[row][2].bar(['no sampling'], [accuracy_no_sampling], color='b')
subs_metrics[row][2].bar(['over sampling'], [accuracy_over_sample], color='orange')
subs_metrics[row][2].bar(['under sampling'], [accuracy_under_sample], color='g')
subs_metrics[row][2].legend([
"no sampling - {:.0%}".format(accuracy_no_sampling),
"over sampling - {:.0%}".format(accuracy_over_sample),
"under sampling - {:.0%}".format(accuracy_under_sample)
])
# Add model name to plots row
subs_metrics[row][0].set_ylabel(model_name)
subs_confusion[row][0].set_ylabel(model_name)
# Clear un-needed figures to coserve memory
for fig_num in plt.get_fignums():
if fig_confusion.number != fig_num and fig_metrics.number != fig_num:
plt.close(fig_num)
# Save model result to file
pickle.dump(model_no_sampling, open('./models/' + model_name.lower().replace(' ', '_') + '_no_sampling', 'wb'))
pickle.dump(model_over_sampling, open('./models/' + model_name.lower().replace(' ', '_') + '_over_sampling', 'wb'));
pickle.dump(model_under_sampling, open('./models/' + model_name.lower().replace(' ', '_') + '_under_sampling', 'wb'));
row += 1
print("====================\n\n")
plt.show()