trainning.py

import numpy as np
import scipy.io as sio
from sklearn import svm
from sklearn.model_selection import GridSearchCV
import pickle

#-load training data from matlab
X = sio.loadmat('spamTrain.mat')['X']
y = sio.loadmat('spamTrain.mat')['y']
Xtest = sio.loadmat('spamTest.mat')['Xtest']
ytest = sio.loadmat('spamTest.mat')['ytest']


#building svm model with grid research
parameter_candidates = [
  {'C': [1, 10, 100, 1000], 'kernel': ['linear']},
  {'C': [1, 10, 100, 1000], 'gamma': [0.001, 0.0001], 'kernel': ['rbf']},
]
# Create a classifier with the parameter candidates
clf = GridSearchCV(estimator=svm.SVC(), param_grid=parameter_candidates, n_jobs=-1)
# Train the classifier on training data
clf.fit(X, np.ravel(y))


# Print out the results
print('Best score for training data:', clf.best_score_)
print('Best `C`:',clf.best_estimator_.C)
print('Best kernel:',clf.best_estimator_.kernel)
print('Best `gamma`:',clf.best_estimator_.gamma)
print('Accuracy is ',clf.score(Xtest,np.ravel(ytest)))

# save model
filename  = "./model/svm_model.pkl"
pickle.dump(clf, open(filename, 'wb'))