def main():
opt = get_opt()
print(opt)
print("GMM: Start to %s, named: %s!" % (opt.stage, "GMM"))
# dataset setup
dataset = Dataset(opt, "GMM")
dataset_loader = DataLoader(opt, dataset)
model = GMM(opt)
if opt.stage == 'train':
if not opt.checkpoint == '' and os.path.exists(opt.checkpoint):
load_checkpoint(model, opt.checkpoint)
train_gmm(opt, dataset_loader, model)
save_checkpoint(
model, os.path.join(opt.checkpoint_dir, opt.name,
'gmm_trained.pth'))
elif opt.stage == 'test':
load_checkpoint(model, opt.checkpoint)
with torch.no_grad():
test_gmm(opt, dataset_loader, model)
else:
raise NotImplementedError('Please input train or test stage')
print('Finished %s stage, named: %s!' % (opt.datamode, opt.name))
def fit_retry(k, data, variant, init, attempt=1):
if attempt >= 100:
print("Failed!")
g = GMM(k)
g.likelihood = 0.0
return g
try:
if init == 'kmeans':
kmeans = KMeans(k)
centroids, labels = kmeans.fit(data)
gmm = GMM(k, mu=centroids, variant=variant,
progress=False, threshold=1e-3)
else:
gmm = GMM(k, variant=variant, progress=False, threshold=1e-3)
gmm.fit(data)
return gmm
except ValueError as e:
return fit_retry(k, data, variant, init, attempt=attempt + 1)
def gmm_log_plot(d, type="full", title=""):
ks, ll = [], []
for i in xrange(1, 5):
data = dataset.read_data(d)
if type == "kmeans":
gmm = GMM(i, mu=KMeans(i).fit(data)[0])
elif type == "diag":
gmm = GMM(i, variant="diag")
else:
gmm = GMM(i)
gmm.fit(data)
ks.append(i)
ll.append(-gmm.likelihood)
print("Likelihood for k = {} => {}".format(i, gmm.likelihood))
plot_loglikelihood(ks, ll, label=type, title=title)
pl.ylabel("Log Likelihood")
pl.xlabel("Number of mixtures")
pl.draw()
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--height', '-ht', type=int, default=32)
parser.add_argument('--width', '-wd', type=int, default=32)
parser.add_argument('--channel', '-ch', type=int, default=1)
parser.add_argument('--batch_size', '-bs', type=int, default=64)
parser.add_argument('--nb_epoch', '-e', type=int, default=1000)
parser.add_argument('--latent_dim', '-ld', type=int, default=2)
parser.add_argument('--save_steps', '-ss', type=int, default=1)
parser.add_argument('--logdir', '-log', type=str, default="../logs")
parser.add_argument('--upsampling', '-up', type=str, default="deconv")
parser.add_argument('--downsampling', '-down', type=str, default="stride")
parser.add_argument('-lr', '--learning_rate', type=float, default=1e-4)
parser.add_argument('--naef', '-naef', type=int, default=16)
parser.add_argument('--distances', '-d', nargs='+', default=['mse'])
parser.add_argument('--nb_components', '-c', type=int, default=4)
args = parser.parse_args()
args_to_csv(os.path.join(args.logdir, 'config.csv'), args)
train_x = get_digits([0, 1], [4000, 10])
image_sampler = ImageSampler(
target_size=(args.width, args.height),
color_mode='gray' if args.channel == 1 else 'rgb',
is_training=True).flow(train_x, batch_size=args.batch_size)
nb_features = args.latent_dim + len(args.distances)
autoencoder = AutoEncoder((args.height, args.width, args.channel),
latent_dim=args.latent_dim,
first_filters=args.naef,
downsampling=args.downsampling,
upsampling=args.upsampling,
distances=args.distances)
estimator = EstimationNetwork((nb_features, ),
dense_units=[256, args.nb_components])
gmm = GMM(args.nb_components, nb_features)
dagmm = DAGMM(autoencoder, estimator, gmm)
dagmm.fit(image_sampler,
nb_epoch=args.nb_epoch,
save_steps=args.save_steps,
logdir=args.logdir)
def gmm_with_kmeans(k, data):
data = dataset.read_data(data)
kmeans = KMeans(k)
centroids, labels = kmeans.fit(data)
# plotKMeans(data, kmeans.mu, labels)
# print centroids.shape
gmm = GMM(k, mu=centroids)
plotMOG(data, params(gmm.fit(data)), title="GMM with KMeans likelihood={}".format(gmm.likelihood))
gmm = GMM(k)
plotMOG(data, params(gmm.fit(data)),
title="GMM general likelihood={}".format(gmm.likelihood))
pl.show()
def gmm_test(k, variant, data_set, title):
gmm = GMM(k, variant=variant)
data = dataset.read_data(name=data_set)
res = gmm.fit(data)
plotMOG(data, params(res, variant), title=title + " Likelihood={}".format(gmm.likelihood))
print "Log Likelihood: ", str(gmm.likelihood)
# print(counter)
# if counter == 100:
# break
kf = KFold(n_splits=3, shuffle=True)
puritites = []
pars = [10, 30, 40, 50, 60]
for par in pars:
print(par)
r = []
pca = PCA(n_components=par)
w = pca.fit_transform(hogs)
for train_index, test_index in kf.split(w):
print("TRAIN:", train_index, "TEST:", test_index)
hog_temp_train, hog_temp_test = w[train_index], w[test_index]
y_temp_train, y_temp_test = train_labels[
train_index], train_labels[test_index]
gmm = GMM(n_components=10, max_iter=120)
gmm.fit(hog_temp_train)
predicted = gmm.predict(hog_temp_test)
acc = purity_score(predicted, y_temp_test)
r.append(acc)
print("total purity: {}".format(acc))
r = np.array(r)
puritites.append(r.mean())
print("after")
acc_max_arg = np.argmax(puritites)
print("best {}: {}".format(pars[acc_max_arg], puritites[acc_max_arg]))
for a, c in zip(puritites, pars):
print("{}: {}".format(c, a))
(train_imgs.shape[0], train_imgs.shape[1] * train_imgs.shape[2]))
# train_imgs = train_imgs +128
# model = NMF(n_components=20, init='random', random_state=0, verbose=True)
# W = model.fit_transform(X=train_imgs)
# H = model.components_
#
# # test_imgs, test_labels = load([i for i in range(10)],'testing')
# W = np.array(W)
#
# print(W.shape)
train_labels = train_labels.reshape(60000)
label_gmm = []
# gmm = mixture.GaussianMixture(n_components=10, verbose=True, max_iter=120)
gmm = GMM(n_components=10, max_iter=120)
counter = 1
while True:
pca = PCA(n_components=counter)
W = pca.fit_transform(train_imgs)
gmm.fit(W)
l = gmm.predict(W)
# for i in range(train_dss.shape[0]):
# label_temp = []
# for j in range(len(train_dss[i])):
# label_temp.append(l[counter])
# counter += 1
# temp = np.argmax(np.bincount(label_temp))
# label_gmm.append(temp)
# label_gmm = np.array(label_gmm)