def _gllim_step(cont: context.abstractHapkeModel, current_noise_cov, current_noise_mean, current_theta):
ti = time.time()
gllim = jGLLiM(K, sigma_type="full", stopping_ratio=stoppingRatioGLLiM)
Xtrain, Ytrain = cont.get_data_training(Ntrain)
Ytrain = cont.add_noise_data(Ytrain, covariance=current_noise_cov, mean=current_noise_mean)
gllim.fit(Xtrain, Ytrain, current_theta, maxIter=maxIterGlliM)
gllim.inversion()
logging.debug(f"GLLiM step done in {time.time() -ti:.3f} s")
return gllim
def interface_julia():
c = context.HapkeContext()
X, Y = c.get_data_training(100000)
gllim = jGLLiM(100, sigma_type="full")
gllim.init_fit(X, Y, "kmeans")
scipy.io.savemat(PATH_OUTPUT, dict(gllim.dict_julia, X=X, Y=Y))
gllim.fit(X, Y, {"rnk": gllim.rnk}, maxIter=10)
scipy.io.loadmat(PATH_OUTPUT)
def setup_jGLLiM_GLLiM():
X = np.random.multivariate_normal(np.zeros(5) + 0.2, np.eye(5), 100000)
Y = np.random.multivariate_normal(np.zeros(6) + 10, np.eye(6), 100000)
gllim = GLLiM(100, 0, sigma_type="full", gamma_type="full", verbose=None)
gllim.init_fit(X, Y, None)
jgllim = jGLLiM(100, 0, sigma_type="full", gamma_type="full", verbose=None)
jgllim.init_fit(X, Y, None)
return gllim, jgllim, X, Y
def _init(cont: context.abstractHapkeModel):
gllim = jGLLiM(K, sigma_type="full", verbose=False)
Xtrain, Ytrain = cont.get_data_training(Ntrain)
Ytrain = cont.add_noise_data(Ytrain, covariance=INIT_COV_NOISE, mean=INIT_MEAN_NOISE) # 0 offset
m = cont.get_X_uniform(K)
rho = np.ones(gllim.K) / gllim.K
precisions = init_X_precision_factor * np.array([np.eye(Xtrain.shape[1])] * gllim.K)
rnk = gllim._T_GMM_init(Xtrain, 'random',
weights_init=rho, means_init=m, precisions_init=precisions)
gllim.fit(Xtrain, Ytrain, {"rnk": rnk}, maxIter=1)
return gllim.theta
def equivalence_jGLLiM_GLLIM():
h = HapkeContext((0, 1, 2, 3))
X = h.get_X_sampling(2000)
Y = np.random.multivariate_normal(np.zeros(10) + 0.4, np.eye(10), 2000)
gllim = GLLiM(10, 0, sigma_type="full", gamma_type="full", verbose=False)
gllim.init_fit(X, Y, "random")
rnk = gllim.rnk
init = {"rnk": rnk}
gllim.fit(X, Y, init, maxIter=5)
jgllim = jGLLiM(10, 0, sigma_type="full", gamma_type="full", verbose=False)
jgllim.fit(X, Y, init, maxIter=5)
theta = (gllim.pikList, gllim.ckList, gllim.GammakList, gllim.AkList,
gllim.bkList, gllim.full_SigmakList)
jtheta = (jgllim.pikList, jgllim.ckList, jgllim.GammakList, jgllim.AkList,
jgllim.bkList, jgllim.full_SigmakList)
is_egal(theta, jtheta)