/
train_svm.py
166 lines (124 loc) · 5.67 KB
/
train_svm.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
import os
import pickle
import pandas as pd
import numpy as np
import train_data
import preprocess as prep
from sklearn.svm import SVC
from sklearn.model_selection import RandomizedSearchCV, GridSearchCV
from termcolor import colored
from evaluate import evaluate
####NOTE: comment out 1/2 of data in train_data.py before running this code
# Generate svc.sav and save it to model folder
def generate_svc(train_df, test_df):
file = open('./eval_results/svc_train_stats.txt', 'w+')
sample_svc = SVC(random_state=40)
print(colored("[SVM]", "green"), " List of parameters:")
for key in sample_svc.get_params().keys():
print("\t- " + str(key))
print(colored("[SVM]", "green"), " Hyperparameters to tune:")
for hyperparam in ['degree', 'kernel', 'gamma']:
print("\t- " + hyperparam)
print("Creating parameter grid for random search ...")
#Degree of the polynomial kernel function (‘poly’)
degree = [2,4,8]
# Specifies the kernel type to be used in the algorithm
kernel = ['poly','rbf']
# Maximum number of levels in tree
gamma = [0.1,0.3,0.5,0.7]
random_grid = {'degree': degree,
'kernel': kernel,
'gamma': gamma}
print(colored("[SVM]", "green"), " Random grid created:")
file.write("Random Grid for Random Search:\n")
for key in random_grid.keys():
to_print = "\t- " + key + ":\t" + str(random_grid[key])
print(to_print)
file.write(to_print + "\n")
file.write("\n")
print(colored("[SVM]", "green"), " Instatiating Random Search ...")
# Base model to tune
clf = SVC()
clf_random = RandomizedSearchCV(estimator=clf, param_distributions=random_grid, n_iter=5, cv=3, verbose=2, random_state=37, n_jobs=-1)
X_train, y_train = prep.get_X_y(train_df)
X_test, y_test = prep.get_X_y(test_df)
print(colored("[SVM]", "green"), " Start Random Search fitting ...")
clf_random.fit(X_train, y_train)
print(colored("[SVM]", "green"), " Random Search fitting DONE with best hyperparameters:")
file.write("Best hyperparameters from Random Search:\n")
bp = clf_random.best_params_
for key in bp.keys():
to_print = "\t- " + key + ":\t" + str(bp[key])
print(to_print)
file.write(to_print + "\n")
file.write("\n")
print(colored("[SVM]", "green"), " Initializing base model ...")
base_model = SVC(degree=bp['degree'],
kernel=bp['kernel'],
gamma=bp['gamma'])
base_model.fit(X_train, y_train)
print(colored("[SVM]", "green"), " Evaluating base model ...")
base_score = evaluate(base_model, X_train, y_train, X_test, y_test, 'SVC', 1)
file.write("Base model evaluation: [refer to svc_eval1.txt]\n\n")
print(colored("[SVM]", "green"), " Creating parameter grid for grid search ...")
# Number of trees in random forest
degree = [bp['degree'],bp['degree']+1]
# Number of features to consider at every split
kernel = [bp['kernel']]
# Maximum number of levels in tree
gamma = [bp['gamma']-0.05,bp['gamma']+0.05,bp['gamma']+0.1]
param_grid = {'degree': degree,
'kernel': kernel,
'gamma': gamma}
print(colored("[SVM]", "green"), " Random grid created:")
file.write("Random Grid for Grid Search:\n")
for key in param_grid.keys():
to_print = "\t- " + key + ":\t" + str(param_grid[key])
print(to_print)
file.write(to_print + "\n")
file.write("\n")
# Base model to tune
clf = SVC()
clf_grid = GridSearchCV(estimator=clf, param_grid=param_grid, cv=3, n_jobs=-1, verbose=2)
print(colored("[SVM]", "green"), " Start Grid Search fitting ...")
clf_grid.fit(X_train, y_train)
print(colored("[SVM]", "green"), " Grid Search fitting DONE with best hyperparameters:")
file.write("Best hyperparameters from Grid Search:\n")
bp = clf_grid.best_params_
for key in bp.keys():
to_print = "\t- " + key + ":\t" + str(bp[key])
print(to_print)
file.write(to_print + "\n")
file.write("\n")
print(colored("[SVM]", "green"), " Initializing best grid model ...")
grid_model = SVC(degree=bp['degree'],
kernel=bp['kernel'],
gamma=bp['gamma'])
grid_model.fit(X_train, y_train)
print(colored("[SVM]", "green"), " Evaluating grid model ...")
grid_score = evaluate(grid_model, X_train, y_train, X_test, y_test, 'SVC', 2)
file.write("Fine tuned model evaluation: [refer to svc_eval2.txt]\n\n")
improvement = 100 * (grid_score - base_score) / base_score
print(colored("[SVM]", "green"), " Improvement of " + str(improvement) + "% from the base model")
file.write("Improvement of " + str(improvement) + "% from the base model\n\n")
if improvement <= 0:
to_print = "Base model is used"
print(colored("[SVM]", "green"), " " + to_print)
file.write(to_print + "\n")
clf = base_model
else:
to_print = "Grid model is used"
print(colored("[SVM]", "green"), " " + to_print)
file.write(to_print + "\n")
clf = grid_model
print(colored("[SVM]", "green"), " Saving SVCmodel ...")
with open('./models/svm.sav', 'wb') as f:
pickle.dump(clf, f)
print(colored("[RF]", "green"), " SVC model saved")
file.close()
if __name__ == "__main__":
train_df, test_df = prep.split_train_test(train_data.load())
print(colored("[CHECK]", "magenta"), " Total train data:\t" + str(len(train_df)))
print(colored("[CHECK]", "magenta"), " Total test data:\t" + str(len(test_df)))
generate_svc(train_df, test_df)
print(colored("[END]", "green"), " SVC training completed")