basemask = np.array(range(1, 25))
svmVectors = []
basemask = basemask - 1
for M in range(MIN_DECIMATION, MAX_DECIMATION + 1):
oDataSet = DataSet()
base = np.loadtxt(PATH_TO_SAVE_FEATURES + "FEATURES_M{}_CM8b_TH199.txt".format(M), usecols=basemask, delimiter=",")
classes = np.loadtxt(PATH_TO_SAVE_FEATURES + "FEATURES_M{}_CM8b_TH199.txt".format(M), dtype=object, usecols=24,
delimiter=",")
for x, y in enumerate(base):
oDataSet.add_sample_of_attribute(np.array(list(np.float32(y)) + [classes[x]]))
oDataSet.attributes = oDataSet.attributes.astype(float)
oDataSet.normalize_data_set()
for j in range(NUMBER_OF_ROUNDS):
print(j)
oData = Data(4, 50, samples=150)
oData.random_training_test_per_class()
svm = ml.SVM_create()
svm.setKernel(ml.SVM_RBF)
oData.params = dict(kernel_type=ml.SVM_RBF, svm_type=ml.SVM_C_SVC, gamma=2.0, nu=0.0, p=0.0, coef0=0,
k_fold=10)
svm.trainAuto(np.float32(oDataSet.attributes[oData.Training_indexes]), ml.ROW_SAMPLE,
np.int32(oDataSet.labels[oData.Training_indexes]), kFold=10)
# svm.train_auto(np.float32(oDataSet.attributes[oData.Training_indexes]),
# np.float32(oDataSet.labels[oData.Training_indexes]), None, None, params=oData.params)
svmVectors.append(svm.getSupportVectors().shape[0])
results = [] # svm.predict_all(np.float32(oDataSet.attributes[oData.Testing_indexes]))
for i in (oDataSet.attributes[oData.Testing_indexes]):
res, cls = svm.predict(np.float32([i]))
results.append(cls[0])
oData.set_results_from_classifier(results, oDataSet.labels[oData.Testing_indexes])
base = np.loadtxt(PATH_TO_SAVE_FEATURES +
"FEATURES_M{}_CM8b_TH{}.txt".format(M, TH),
usecols=basemask,
delimiter=",")
classes = np.loadtxt(PATH_TO_SAVE_FEATURES +
"FEATURES_M{}_CM8b_TH{}.txt".format(M, TH),
dtype=object,
usecols=24,
delimiter=",")
for x, y in enumerate(base):
oDataSet.add_sample_of_attribute(
np.array(list(np.float32(y)) + [classes[x]]))
oDataSet.attributes = oDataSet.attributes.astype(float)
oDataSet.normalize_data_set()
for j in range(ROUNDS):
oData = Data(4, 13, samples=50)
oData.random_training_test_per_class()
svm = cv2.SVM()
oData.params = dict(kernel_type=cv2.SVM_RBF,
svm_type=cv2.SVM_C_SVC,
gamma=2.0,
nu=0.0,
p=0.0,
coef0=0,
k_fold=2)
svm.train_auto(np.float32(oDataSet.attributes[oData.Training_indexes]),
np.float32(oDataSet.labels[oData.Training_indexes]),
None,
None,
params=oData.params)
oData.insert_model(svm)
[2.88141800e+01, 7.24622100e-02],
[9.68741850e-02, -8.40647000e-02],
[4.80887420e+04, 5.70339160e+00]])
basemask = np.array([1, 2, 5, 9, 15, 16, 17, 21, 22, 23])
basemask = basemask - 1
svm = cv2.SVM()
oDataSet = DataSet()
base = np.loadtxt(PATH_TO_SAVE_FEATURES +
"FEATURES_M{}_CM8b_TH198.txt".format(M),
usecols=basemask,
delimiter=",")
classes = np.loadtxt(PATH_TO_SAVE_FEATURES +
"FEATURES_M{}_CM8b_TH198.txt".format(M),
dtype=object,
usecols=24,
delimiter=",")
for x, y in enumerate(base):
oDataSet.add_sample_of_attribute(
np.array(list(np.float32(y)) + [classes[x]]))
oDataSet.attributes = oDataSet.attributes.astype(float)
oDataSet.attributes = (oDataSet.attributes - NORMALIZATION_MATRIX[:, 1].T) / (
NORMALIZATION_MATRIX[:, 0] - NORMALIZATION_MATRIX[:, 1])
oData = Data(4, 13, samples=50)
svm.load("MODEL_M1_CM8_TH198_ATT10_ROUND_24.txt")
results = svm.predict_all(np.float32(oDataSet.attributes))
oData.set_results_from_classifier(results, oDataSet.labels)
oData.insert_model(svm)
print oData.confusion_matrix
print oData
usecols=basemask,
delimiter=",")
classes = np.loadtxt(
PATH_TO_SAVE_FEATURES +
"FEATURES_M{}_CM{}b_TH198.txt".format(DECIMATION, n_bits),
dtype=object,
usecols=24,
delimiter=",")
for x, y in enumerate(base):
oDataSet.add_sample_of_attribute(
np.array(list(np.float32(y)) + [classes[x]]))
oDataSet.attributes = oDataSet.attributes.astype(float)
oDataSet.normalize_data_set()
for j in range(NUMBER_OF_ROUNDS):
print j
oData = Data(4, 13, samples=50)
oData.random_training_test_per_class()
svm = cv2.SVM()
oData.params = dict(kernel_type=cv2.SVM_RBF,
svm_type=cv2.SVM_C_SVC,
gamma=2.0,
nu=0.0,
p=0.0,
coef0=0,
k_fold=2)
svm.train_auto(np.float32(oDataSet.attributes[oData.Training_indexes]),
np.float32(oDataSet.labels[oData.Training_indexes]),
None,
None,
params=oData.params)
oData.insert_model(svm)