def main():
y_train, x_train = svm_read_problem('../Data/hw5/satimage.scale')
y_train = [temp == 3.0 for temp in y_train]
model = svm_train(y_train, x_train, '-t 0 -c 10')
sv_coef = np.array(model.get_sv_coef())
sv_indices = [index - 1 for index in model.get_sv_indices()]
_, x_support = svm_read_problem('../Data/hw5/satimage.scale',
return_scipy=True)
x_support = x_support.toarray()[sv_indices]
print(f'||w|| is {np.linalg.norm(x_support.T @ sv_coef)}')
def LoadLibSVMData(path: str,
features: int,
x_dtype: np.dtype = np.float,
y_dtype: np.dtype = np.int) -> np.ndarray:
"""
加载 LibSVM 数据集的数据，转成numpy.ndarray类型
@Param:
-------
path: 文件路径\n
features: 数据维度
@Return:
--------
data_x: [numpy.ndarray](n_samples, n_features)
data_y: [numpy.ndarray](n_samples,)
"""
libsvm_data = None
try:
libsvm_data = svm_read_problem(path)
except IOError:
raise ("文件路径出错")
libsvm_y, libsvm_x = libsvm_data
data_y = np.array(libsvm_y, dtype=y_dtype)
data_x = np.zeros((len(libsvm_x), features), dtype=x_dtype)
for line, map_ in zip(data_x, libsvm_x):
for pair in map_.items():
line[pair[0] - 1] = pair[1]
return (data_x, data_y)
def test_v_svr(prompt_idx, gamma=None):
args = _parse_args()
dataset = AESDataset(args.dataset_dir,
prompt_num=PROMPT_NUM,
is_cross_dataset=False)
x_train_list, y_train_list = dataset.get_train()
x_dev_list, y_dev_list = dataset.get_dev()
x_test_list, _, _ = dataset.get_test()
train_len, dev_len, test_len = len(x_train_list[prompt_idx-1]), len(x_dev_list[prompt_idx-1]),\
len(x_test_list[prompt_idx-1])
y, x = svm_read_problem(SVM_SCALE_DIR + '/prompt@' + str(prompt_idx) +
'-scale.txt')
x_train, y_train = x[:train_len], y[:train_len]
x_dev, y_dev = x[train_len:train_len + dev_len], y[train_len:train_len +
dev_len]
x_test = x[train_len + dev_len:]
if gamma:
param = f'-s 4 -t 2 -c 1000 -n 0.1 -g {gamma}'
else:
param = f'-s 4 -t 2 -c 1000 -n 0.1'
svm_model = svm_train(y_train + y_dev, x_train + x_dev, param)
p_label, p_acc, p_val = svm_predict(np.zeros(shape=len(x_test)), x_test,
svm_model)
p_label = np.round(p_label)
dev_label, dev_acc, dev_val = svm_predict(y_dev, x_dev, svm_model)
dev_kappa = kappa(y_true=y_dev, y_pred=dev_label, weights='quadratic')
print(f'Dev kappa: {dev_kappa}')
return dev_kappa, p_label
def main():
y_train, x_train = svm_read_problem('../Data/hw5/satimage.scale')
y_train = [temp == 6.0 for temp in y_train]
y_test, x_test = svm_read_problem('../Data/hw5/satimage.scale.t')
y_test = [temp == 6.0 for temp in y_test]
for c in [0.01, 0.1, 1, 10, 100]:
print(f'---------')
print(f'C {c}')
model = svm_train(y_train, x_train, f'-t 2 -g 10 -c {c}')
p_label, p_acc, p_val = svm_predict(y_test, x_test, model)
print(f'Eout = {p_acc[1]}')
def main():
y_train, x_train = svm_read_problem('../Data/hw5/satimage.scale')
for i in range(5):
print(f'Label {i+1}')
temp_y_train = [temp == i+1 for temp in y_train]
model = svm_train(temp_y_train, x_train, '-t 1 -d 2 -g 1 -r 1 -c 10')
p_label, p_acc, p_val = svm_predict(temp_y_train, x_train, model)
def main():
y_train, x_train = svm_read_problem('../Data/hw5/satimage.scale',
return_scipy=True)
y_train = y_train == 6.0
gammas = {g: 0 for g in [0.1, 1, 10, 100, 1000]}
with ProcessPoolExecutor() as executor:
for _, idx in tqdm(executor.map(one_iter, [y_train] * ITER,
[x_train] * ITER),
total=ITER):
gammas[idx] += 1
print(gammas)
def heart_scale():
return svm_read_problem(os.path.join(TEST_RESOURCES_PATH, 'heart_scale'))
def get_data(path: str):
root_dir = os.path.dirname(os.path.abspath(__file__))
data = commonutil.svm_read_problem(os.path.join(root_dir, path), True)
return np.insert(data[1].A, 0, np.arange(1,
len(data[0]) + 1),
axis=1), data[0]