def get_bgm(n_clusters=10, wcp=1.0e3, tol=None):
"""
# wcp: higher number puts more mass in the center and will lead to more
# more components being active
"""
covartype = 'full'
#init_params = 'random' #alternative is kmeans
init_params = 'kmeans' #alternative is kmeans
#wcpt = 'dirichlet_process' #wcp shouldbe (float, float)
wcpt = 'dirichlet_distribution' #wcp should be float
warm_start = False
verbose = False
n_init = 1
max_iter = 500
reg_covar = 1e-5 #default is 1e-6
if not tol: tol = 1e-3
gmm = BGM(n_components=n_clusters,
covariance_type=covartype,
n_init=n_init,
weight_concentration_prior_type=wcpt,
init_params=init_params,
max_iter=max_iter,
verbose=verbose,
reg_covar=reg_covar,
tol=tol)
return gmm
def _bgm_fit(self, x):
"""Fit a Bayesian Gaussian Mixture to the data given by x.
Parameters
----------
x : array-like, shape (n_samples, n_attributes)
The data to be fit.
Returns
-------
model : BayesianGaussianMixture from the sklearn package
The BayesianGaussianMixture object that has been fit to the data.
"""
model = BGM(n_components=self.n_components,
tol=self.tol,
max_iter=self.max_iter,
n_init=self.n_init,
covariance_type=self.cov_type,
weight_concentration_prior_type=self.
weight_concentration_prior_type,
weight_concentration_prior=self.weight_concentration_prior)
data = x.astype('float32')
model.fit(data)
return model
def fit(self, X, Y):
# assume classes are numbered 0...K-1
self.K = len(set(Y))
self.gaussians = []
self.p_y = np.zeros(self.K)
for k in range(self.K):
print("Fitting gmm", k)
Xk = X[Y == k]
self.p_y[k] = len(Xk)
gmm = BGM(10)
gmm.fit(Xk)
self.gaussians.append(gmm)
# normalize p(y)
self.p_y /= self.p_y.sum()
def return_clustering(self, time_step, n_clusters=10):
c_model = BGM(n_clusters, covariance_type='diag').fit(self.x_train)
self.clustering = c_model
if self.count % self.trading_window == 0:
X_principal = pd.DataFrame(self.PCA(time_step, 2)[0])
X_principal.columns = ['P1', 'P2']
X_principal['clusters'] = self.clustering.predict(self.x_train)
sns.pairplot(x_vars=["P1"],
y_vars=["P2"],
data=X_principal,
hue="clusters",
size=5)
plt.show()
self.count += 1
def copy_fit(bgm, method='bgm'):
n_clusters = bgm.n_components
covartype = bgm.covariance_type
n_init = bgm.n_init
max_iter = bgm.max_iter
tol = bgm.tol
verbose = True
if method == 'bgm':
wcpt = bgm.weight_concentration_prior_type
reg_covar = bgm.reg_covar
init_params = bgm.init_params
tol = bgm.tol
copy = BGM(n_components=n_clusters,
covariance_type=covartype,
n_init=n_init,
weight_concentration_prior_type=wcpt,
init_params=init_params,
max_iter=max_iter,
verbose=verbose,
reg_covar=reg_covar,
tol=tol)
copy.weight_concentration_prior_ = bgm.weight_concentration_prior_
copy.weight_concentration_ = bgm.weight_concentration_
copy.mean_precision_prior = bgm.mean_precision_prior
copy.mean_prior_ = bgm.mean_prior_
copy.mean_precision_ = bgm.mean_precision_
copy.covariance_prior_ = bgm.covariance_prior_
copy.degrees_of_freedom_prior_ = bgm.degrees_of_freedom_prior_
copy.degrees_of_freedom_ = bgm.degrees_of_freedom_
if method == 'gmm':
copy = GMM(n_components=n_clusters,
random_state=42,
covariance_type=covartype,
max_iter=max_iter,
n_init=n_init,
tol=tol,
verbose=verbose)
copy.means_ = bgm.means_
copy.covariances_ = bgm.covariances_
copy.weights_ = bgm.weights_
copy.precisions_ = bgm.precisions_
copy.precisions_cholesky_ = bgm.precisions_cholesky_
copy.converged_ = bgm.converged_
copy.n_iter_ = bgm.n_iter_
copy.lower_bound_ = bgm.lower_bound_
return copy
# [[[ 0.35313405 -0.06505064] # [-0.06505064 0.27590391]] # [[ 0.46521907 0.08082033] # [ 0.08082033 0.34744874]] # [[ 0.40641319 -0.07908257] # [-0.07908257 0.20770555]]] # print(gm.converged_) # True # print(gm.n_iter_) # 5 # print(gm.predict(X)) # 硬分群 # print(gm.predict_proba(X)) # 軟分群 # 高斯混合模型是生成模型, 也就是可以從裡面抽樣新實例 # import numpy as np # X_new, y_new = gm.sample(6) # densities = gm.score_samples(X) # 估算模型在任何位置的密度, 此方法可以估計它收到的實例位置的機率密度函數(PDF) # density_threshold = np.percentile(densities, 4) # anomalies = X[densities < density_threshold] # gm.bic(X) # gm.aic(X) # 貝氏高斯混合模型 from sklearn.mixture import BayesianGaussianMixture as BGM import numpy as np bgm = BGM(n_components=10, n_init=10) bgm.fit(X) print(np.round(bgm.weights_, 2))
Tprocess0 = time.time()
print('\n', '## DATE PREPARATION RUNTIME:', Tprocess0 - Tstart) #Timer
## MAIN ##
#load CAE model
cae_model = load_model(cae_mfn)
#Retrieve the ecoder layer
Embedding_layer = K.function([cae_model.layers[0].input],
[cae_model.layers[14].output])
input4bgmm = Embedding_layer([X_train[:]])
input4bgmm = np.array(input4bgmm)
input4bgmm = input4bgmm[0]
print(input4bgmm.shape)
#clustering
grouper = BGM(n_components=nCluster)
grouper.fit(input4bgmm)
if tosavemodel:
#restore the model
pickle.dump(grouper, open(savename, 'wb'))
Tprocess1 = time.time()
print('\n', '## CLUSTERING RUNTIME:', Tprocess1 - Tprocess0) #Timer end
#brief examination
y_pred = grouper.predict(input4bgmm)
y_max = np.max(y_pred)
y_proba = grouper.predict_proba(
input4bgmm) #probability of being a certain group
#group = [(number of group members): images, group label, probability for each group]