def group_fit(features):
"""
根据特征区分样本是否为噪音标签
:param features:
:return:
"""
model = GaussianMixture(2, max_iter=10, tol=1e-2, reg_covar=5e-4)
model.fit(features)
if np.argmax(model.means_[:,
1]) == 0: # 根据混合分布的均值参数来区分，概率高的将其看做是clean label
parameters = [i[::-1] for i in model._get_parameters()]
model._set_parameters(parameters)
return model
# df[df[:, 0] == 'U'] = 20
# df[df[:, 0] == 'V'] = 21
# df[df[:, 0] == 'W'] = 22
# df[df[:, 0] == 'X'] = 23
# df[df[:, 0] == 'Y'] = 24
# df[df[:, 0] == 'Z'] = 25
# df_target_names = ['A', 'B', 'C']
df = df[:200, :]
df_data = df[:, 1:]
df_target = df[:, 0]
gmm = GaussianMixture(n_components=16, n_init=10)
gmm.fit(df_data)
prob = gmm.predict_proba(df_data)
print(prob.shape)
para = gmm._get_parameters()
print(len(para)) # tuple length
for i in range(len(para)):
print(para[i].shape)
# X_train = df_data[:15000, 1:]
# y_train = df_target[:15000]
# X_test = df_data[15000:, 1:]
# y_test = df_target[15000:]
#
# n_classes = len(np.unique(y_train))
#
# # Try GMMs using different types of covariances.
# estimators = dict((cov_type, GaussianMixture(n_components=n_classes,
# covariance_type=cov_type, max_iter=200, random_state=0))
# for cov_type in ['spherical', 'diag', 'tied', 'full'])
