Esempio n. 1
0
def svm1(X, X_valid, y, Y_valid,train_x,train_y,kaggle_test):
	#X,y-training set
	#X_valid -validationset
	#train_x,whole training set
	#X_test kaggle
	m, n = X.shape
	kf = KFold(m, n_folds=2)
	gammas = [0.001,0.01,0.1]
	cs = [1.0,10.0,100.0]
	accuracy_list = []
	for g in gammas:
		for c in cs:
			accs = []
			counter = 0 
			for train_index, test_index in kf:
				X_train, X_test = X[train_index], X[test_index]
				y_train, y_test = y[train_index], y[test_index]
				clf = svm.SVC(C= c,gamma=g)
				clf.fit(X_train,y_train)
				pred = clf.predict(X_test)
				# print type(pred)
				corr, acc = accuracy(y_test,pred)
				accs.append(acc)
				counter +=1
			acc_mean = np.mean(accs)
			accuracy_list.append((c,g,acc_mean))

	sortedaccuracy_list = sorted(accuracy_list, key = lambda x:x[2], reverse = True)
	print sortedaccuracy_list
	c_optimal = sortedaccuracy_list[0][0]
	g_optimal = sortedaccuracy_list[0][1]
	fname = 'results/svm/svm_accuracy.png'
	plot_svm_accuracy(accuracy_list,fname)

	# check validationset
	print "testing on validationset"
	clf = svm.SVC(C= c_optimal,gamma=g_optimal)
	clf.fit(X,y)
	pred__valid_svm = clf.predict(X_valid)
	print "results of svm using "+str(c_optimal)+'c'+str(g_optimal)+'gamma' + ":\n%s\n" %(metrics.classification_report(Y_valid,pred__valid_svm))
	fname2 = 'results/svm/'+str(c_optimal)+'c'+str(g_optimal)+'gamma'+'cfm_svm.png'
	cfm_svm(pred__valid_svm,Y_valid,c_optimal,g_optimal,fname2)
	# classifcation of kaggle testset
	print "classifcation kaggle"
	clf = svm.SVC(C= c_optimal,gamma=g_optimal)
	clf.fit(train_x,train_y)
	pred_kaggle_svm = clf.predict(kaggle_test)
	fname3 = 'results/svm/svm_kaggle.csv'
	write_tokaggle(fname3,pred_kaggle_svm)
Esempio n. 2
0
def svm2_pca(X, X_valid, y, Y_valid,train_x,train_y,kaggle_test):
	pcas = [100,500,1000]
	gammas = [0.001,0.01,0.1]
	cs = [1.0,10.0,100.0]
	accuracy_list = []
	for pca in pcas:
		print 'pca',pca
		accuracy_list_p = []
		pca = PCA(n_components = pca,whiten = True)
		train_x_p = pca.fit_transform(train_x)

		X_p = pca.transform(X)
		X_valid_p = pca.transform(X_valid)
		kaggle_test_p = pca.transform(kaggle_test)
		m, n = X_p.shape
		kf = KFold(m, n_folds=2)
		for g in gammas:
			for c in cs:
				accs = []
				counter = 0 
				for train_index, test_index in kf:
					X_train, X_test = X_p[train_index], X_p[test_index]
					y_train, y_test = y[train_index], y[test_index]
					clf = svm.SVC(C= c,gamma=g)
					clf.fit(X_train,y_train)
					pred = clf.predict(X_test)
					# print type(pred)
					corr, acc = accuracy(y_test,pred)
					accs.append(acc)
					counter +=1
				acc_mean = np.mean(accs)
				accuracy_list.append((pca,c,g,acc_mean))
				accuracy_list_p.append((c,g,acc_mean))
		fname = 'results/svm_pca/svm_pca_accuracy_'+str(pca)+'pca.png'
		plot_svm_accuracy(accuracy_list_p,fname)

	sortedaccuracy_list = sorted(accuracy_list, key = lambda x:x[3], reverse = True)
	print sortedaccuracy_list
	pca_optimal = sortedaccuracy_list[0][0]
	c_optimal = sortedaccuracy_list[0][1]
	g_optimal = sortedaccuracy_list[0][2]

	pca = PCA(n_components = pca_optimal,whiten = True)
	train_x_p = pca.fit_transform(train_x)
	X_p = pca.transform(X)
	X_valid_p = pca.transform(X_valid)
	kaggle_test_p = pca.transform(kaggle_test)
	# check validationset
	print "testing on validationset"

	clf = svm.SVC(C= c_optimal,gamma=g_optimal)
	clf.fit(X_p,y)
	pred__valid_svm = clf.predict(X_valid_p)
	print "results of svm using pca "+str(pca_optimal)+'pca components'+str(c_optimal)+'c'+str(g_optimal)+'gamma' + ":\n%s\n" %(metrics.classification_report(Y_valid,pred__valid_svm))
	fname2 = 'results/svm_pca/'+str(pca_optimal)+'pca components'+str(c_optimal)+'c'+str(g_optimal)+'gamma'+'cfm_svm_pca.png'
	cfm_svm(pred__valid_svm,Y_valid,c_optimal,g_optimal,fname2)
	# classifcation of kaggle testset
	print "classifcation kaggle"
	clf = svm.SVC(C= c_optimal,gamma=g_optimal)
	clf.fit(train_x_p,train_y)
	pred_kaggle_svm = clf.predict(kaggle_test_p)
	fname3 = 'results/svm_pca/svm_pca_kaggle.csv'
	write_tokaggle(fname3,pred_kaggle_svm)