def test(self):
D = 3
K = 4
hyper_params = pa.HyperParameters(dim=D, k=K, nu=D * torch.ones(K))
d = Dirichlet(hyper_params.alpha)
s = d.sample()
self.assertTrue(s.size() == (K, ))
def test(self):
N = 2
D = 3
K = 4
eta = torch.arange(N * K, dtype=torch.float32).reshape(N, K)
dataset = torch.arange(N * D, dtype=torch.float32).reshape(N, D)
hyper_params = pa.HyperParameters(dim=D, k=K, nu=D * torch.ones(K))
updater = QpiUpdater(hyper_params)
updater.update(dataset, eta)
self.assertTrue((K, ) == updater.alpha.size())
def test(self):
K = 3
D = 2
N = 4
dataset = torch.arange(N * D, dtype=torch.float32).reshape(N, D)
eta = torch.arange(N * K, dtype=torch.float32).reshape(N, K)
hyper_params = pa.HyperParameters(dim=D, k=K, nu=D * torch.ones(K))
updater = QmuUpdater(hyper_params)
updater.update(dataset, eta)
self.assertTrue(updater.beta.size() == (K, ))
self.assertTrue(updater.m.size() == (K, D))
def test(self):
N = 2
D = 3
K = 4
eta = torch.arange(N * K, dtype=torch.float32).reshape(N, K)
beta = torch.arange(K, dtype=torch.float32)
m = torch.arange(K * D, dtype=torch.float32).reshape(K, D)
dataset = torch.arange(N * D, dtype=torch.float32).reshape(N, D)
hyper_params = pa.HyperParameters(dim=D, k=K, nu=D * torch.ones(K))
updater = QlambdaUpdater(hyper_params)
updater.update(dataset, eta, beta, m)
self.assertTrue((K, D, D) == updater.W.size())
self.assertTrue((K,) == updater.nu.size())
plt.scatter(dataset[:, 0], dataset[:, 1], marker='.', c=colors)
plt.scatter(xs.ravel(), ys.ravel(), marker=".", c=pcolors, alpha=0.1)
plt.xlim(X_MIN, X_MAX)
plt.ylim(Y_MIN, Y_MAX)
plt.savefig('./predict.jpg')
def make_initial_positions_with_kmeans(dataset, k):
p = cl.KMeans(n_clusters=k).fit(dataset)
return p.cluster_centers_
if __name__ == "__main__":
try:
hyper_params = pa.HyperParameters(dim=DIM, k=K, nu=NU)
qs_updater = qs.QsUpdater()
qp_updater = qp.QpiUpdater(hyper_params)
qm_updater = qm.QmuUpdater(hyper_params)
ql_updater = ql.QlambdaUpdater(hyper_params)
dataset = ds.make_dataset_0(OBS_NUM, DIM, K)
std, mean = torch.std_mean(dataset, dim=0)
dataset = (dataset - mean) / std
display_graph(dataset)
cs = make_initial_positions_with_kmeans(dataset, K)
# initialize mu
qm_updater.m = torch.tensor(cs).float()
prev_m = qm_updater.m.clone()