def test_initialize_with_small_n_features():
N = 500
n_components = 128
n_features = 64
X = np.random.randn(N, n_features)
dico = ApproximateKSVD(n_components=n_components)
dico.fit(X)
def get_decomp(input_image):
patches = image.extract_patches_2d(input_image, (16,16))
signals = patches.reshape(patches.shape[0], -1)
aksvd = ApproximateKSVD(n_components=64, max_iter = 100, transform_n_nonzero_coefs=4)
dictionary = aksvd.fit(signals[np.random.choice(range(signals.shape[0]), 1000, replace = True)]).components_
gamma = aksvd.transform(signals)
return gamma, dictionary
def getDict(img_chunk):
patches = image.extract_patches_2d(img_chunk, (16,16))
signals = patches.reshape(patches.shape[0], -1)
# Set K-SVD paprameters to maximize classification rate according to fannjiang
aksvd = ApproximateKSVD(n_components=64, max_iter = 100, transform_n_nonzero_coefs=4)
dictionary = aksvd.fit(signals[np.random.choice(range(signals.shape[0]), 1000, replace = True)]).components_
#gamma = aksvd.transform(signals)
return dictionary
def test_fit():
np.random.seed(0)
N = 1000
L = 256
X = np.random.randn(N, 20) + np.random.rand(N, 20)
dico = ApproximateKSVD(n_components=L)
dico.fit(X)
gamma = dico.transform(X)
assert_array_almost_equal(X, gamma.dot(dico.components_))
def getDict(signals, n):
# Set K-SVD paprameters to maximize classification rate according to fannjiang
print(n)
aksvd = ApproximateKSVD(n_components=64,
max_iter=50,
transform_n_nonzero_coefs=4)
dictionary = aksvd.fit(signals[np.random.choice(range(signals.shape[0]),
int(round(n)),
replace=True)]).components_
#gamma = aksvd.transform(signals)
return dictionary
class KSVD(ClusteringAlgo):
def __init__(self, *args, **kwargs):
self._aksvd = ApproximateKSVD(*args, **kwargs)
def cluster(self, features):
self._aksvd.fit(features)
gamma = self._aksvd.transform(features)
assignments = np.argmax(np.abs(gamma), axis=1)
return assignments
def __str__(self):
return "ksvd"
def test_fit():
np.random.seed(0)
N = 1000
L = 64
n_features = 128
B = np.array(sp.sparse.random(N, L, density=0.5).todense())
D = np.random.randn(L, n_features)
X = np.dot(B, D)
dico = ApproximateKSVD(n_components=L, transform_n_nonzero_coefs=L)
dico.fit(X)
gamma = dico.transform(X)
assert_array_almost_equal(X, gamma.dot(dico.components_))
def class_dict_coder(X, y, n_class_atoms=None, n_class=None):
# For each class do ksvd using library
D = np.zeros((X.shape[0], n_class_atoms * 4))
for i in range(n_class):
#print("Doing dictionary initialization for ", i)
X1 = X[:, y == i]
aksvd = ApproximateKSVD(n_components=n_class_atoms)
dictionary = aksvd.fit(np.transpose(X1)).components_
D[:,
i * n_class_atoms:(i + 1) * n_class_atoms] = np.transpose(dictionary)
#print("Done dictionary initialization for ", i)
return D
def dictionary_KSVD(num_clusters, word_vectors):
# Initalize a ksvd object and use it for clustering.
aksvd = ApproximateKSVD(n_components=num_clusters)
dictionary = aksvd.fit(word_vectors).components_
idx_proba = aksvd.transform(word_vectors)
idx = np.argmax(idx_proba, axis=1)
print "Clustering Done...", time.time() - start, "seconds"
# Get probabilities of cluster assignments.
# Dump cluster assignments and probability of cluster assignments.
joblib.dump(idx, 'ksvd_latestclusmodel_len2alldata.pkl')
print "Cluster Assignments Saved..."
joblib.dump(idx_proba, 'ksvd_prob_latestclusmodel_len2alldata.pkl')
print "Probabilities of Cluster Assignments Saved..."
return (idx, idx_proba)
def cluster_GMM(num_clusters, word_vectors, transform_n_nonzero_coefs=None):
# Initalize a GMM object and use it for clustering.
# clf = GMM(n_components=num_clusters,covariance_type="tied", init_params='wc', n_iter=10)
aksvd = ApproximateKSVD(
n_components=num_clusters,
transform_n_nonzero_coefs=transform_n_nonzero_coefs)
dictionary = aksvd.fit(word_vectors).components_
idx_proba = aksvd.transform(word_vectors)
idx = np.argmax(idx_proba, axis=1)
print("Clustering Done...", time.time() - start, "seconds")
joblib.dump(idx, 'data_files/gmm_latestclusmodel_len2alldata.pkl')
print("Cluster Assignments Saved...")
joblib.dump(idx_proba,
'data_files/gmm_prob_latestclusmodel_len2alldata.pkl')
print("Probabilities of Cluster Assignments Saved...")
return (idx, idx_proba)
def test_size():
np.random.seed(0)
N = 100
L = 128
X = np.random.randn(N, 10) + np.random.rand(N, 10)
dico1 = ApproximateKSVD(n_components=L)
dico1.fit(X)
gamma1 = dico1.transform(X)
e1 = norm(X - gamma1.dot(dico1.components_))
dico2 = DictionaryLearning(n_components=L)
dico2.fit(X)
gamma2 = dico2.transform(X)
e2 = norm(X - gamma2.dot(dico2.components_))
assert dico1.components_.shape == dico2.components_.shape
assert gamma1.shape == gamma2.shape
assert e1 < e2
def ksvd(self):
"""K-SVD denoising algorithm"""
P = extract_patches_2d(self.Inoisy, self.ksvd_patch)
patch_shape = P.shape
P = P.reshape((patch_shape[0], -1))
mean = np.mean(P, axis=1)[:, np.newaxis]
P -= mean
aksvd = ApproximateKSVD(n_components=self.ksvd_components)
dico = aksvd.fit(P).components_
reduced = (aksvd.transform(P)).dot(dico) + mean
reduced_img = reconstruct_from_patches_2d(reduced.reshape(patch_shape),
self.shape)
self.Iksvd = np.clip(reduced_img, 0, 1)
self.Ilist[self.str2int['ksvd']] = self.Iksvd
if self.verbose:
print('K-SVD :', self.Iksvd)
return ()
def dictionary_KSVD(num_clusters, word_vectors, basic_path: str):
# Initalize a ksvd object and use it for clustering.
aksvd = ApproximateKSVD(n_components=num_clusters)
dictionary = aksvd.fit(word_vectors).components_
idx_proba = aksvd.transform(word_vectors)
idx = np.argmax(idx_proba, axis=1)
# print("Clustering Done...", time.time() - start, "seconds")
# Get probabilities of cluster assignments.
# Dump cluster assignments and probability of cluster assignments.
joblib.dump(
idx,
os.path.join(config["system_storage"]["models"], f'ksvd_{basic_path}'))
print("Cluster Assignments Saved...")
joblib.dump(
idx_proba,
os.path.join(config["system_storage"]["models"],
f'ksvd_prob_{basic_path}'))
print("Probabilities of Cluster Assignments Saved...")
return (idx, idx_proba)
def test_size():
np.random.seed(0)
N = 50
L = 12
n_features = 16
D = np.random.randn(L, n_features)
B = np.array(sp.sparse.random(N, L, density=0.5).todense())
X = np.dot(B, D)
dico1 = ApproximateKSVD(n_components=L, transform_n_nonzero_coefs=L)
dico1.fit(X)
gamma1 = dico1.transform(X)
e1 = norm(X - gamma1.dot(dico1.components_))
dico2 = DictionaryLearning(n_components=L, transform_n_nonzero_coefs=L)
dico2.fit(X)
gamma2 = dico2.transform(X)
e2 = norm(X - gamma2.dot(dico2.components_))
assert dico1.components_.shape == dico2.components_.shape
assert gamma1.shape == gamma2.shape
assert e1 < e2
def my_cross_val_score(CZ,
n_comp=3,
model='pca',
k=5,
random_state=False,
shuffle=True):
if shuffle == True:
save = np.copy(CZ)
np.random.shuffle(CZ)
kf = KFold(n_splits=k)
cv = []
pca = decomposition.PCA(n_components=n_comp, random_state=random_state)
nmf = NMF(n_components=n_comp, random_state=random_state)
kmeans = KMeans(n_clusters=n_comp, random_state=random_state)
ksvd = ApproximateKSVD(n_components=n_comp)
for train, test in kf.split(CZ):
cz_test = CZ[test]
cz_train = CZ[train]
if model == 'pca':
pca.fit(cz_train)
CZ_reconstructed = pca.inverse_transform(pca.transform(cz_test))
elif model == 'nmf':
nmf.fit(cz_train)
CZ_reconstructed = np.dot(nmf.transform(cz_test), nmf.components_)
elif model == 'k_means':
kmeans.fit(cz_train)
CZ_reconstructed = kmeans.cluster_centers_[kmeans.predict(cz_test)]
elif model == 'k_svd':
meantr = np.mean(cz_train, axis=1)[:, np.newaxis]
meantest = np.mean(cz_test, axis=1)[:, np.newaxis]
dictionary = ksvd.fit(cz_train - meantr).components_
gamma = ksvd.transform(cz_test - meantest)
CZ_reconstructed = gamma.dot(dictionary) + meantest
cv.append(mean_squared_error(CZ_reconstructed, cz_test))
if shuffle == True:
CZ = save
return cv
def researcher_ksvd(CLF, X_train, Y_train, X_test, Y_test, k=2, n_comp=[3,5,7,10], **kwargs):
scores = []
for n in n_comp:
ksvd = ApproximateKSVD(n_components=n, transform_n_nonzero_coefs=max(n-k, 1))
meantr = np.mean(X_train,axis=0)
ksvd.fit(X_train - meantr).components_
gamma_train = ksvd.transform(X_train - meantr)
gamma_test = ksvd.transform(X_test - meantr)
clf = CLF(**kwargs)
clf.fit(gamma_train, Y_train)
predict = clf.predict(gamma_test)
score = accuracy_score(Y_test, predict)
scores.append(score)
return scores
from skimage import io, util
from sklearn.feature_extraction import image
from sklearn import preprocessing
from ksvd import ApproximateKSVD
parser = argparse.ArgumentParser()
parser.add_argument('--input', type=str, required=True)
parser.add_argument('--output', type=str, required=True)
args = parser.parse_args()
def clip(img):
img = np.minimum(np.ones(img.shape), img)
img = np.maximum(np.zeros(img.shape), img)
return img
img = util.img_as_float(io.imread(args.input))
patch_size = (5, 5)
patches = image.extract_patches_2d(img, patch_size)
signals = patches.reshape(patches.shape[0], -1)
mean = np.mean(signals, axis=1)[:, np.newaxis]
signals -= mean
aksvd = ApproximateKSVD(n_components=32)
dictionary = aksvd.fit(signals[:10000]).components_
gamma = aksvd.transform(signals)
reduced = gamma.dot(dictionary) + mean
reduced_img = image.reconstruct_from_patches_2d(reduced.reshape(patches.shape),
img.shape)
io.imsave(args.output, clip(reduced_img))
def __init__(self, *args, **kwargs):
self._aksvd = ApproximateKSVD(*args, **kwargs)
## Dictionary Learning:
print('Dictionary Learning ...')
R = np.random.randn(1728, 128)
n_components = 128 * 2 # Over complete factor = 2
transform_n_nonzero_coefs = 12
for Y, label in zip(Descr, weizmann_label_dic):
# Y subtract mean ????????
Y = np.dot(Y, R) # Y: k x 128
aksvd = ApproximateKSVD(
n_components=n_components,
transform_n_nonzero_coefs=transform_n_nonzero_coefs)
D = aksvd.fit(Y).components_
X = aksvd.transform(Y)
print('YShape:', Y.shape)
print('DShape:', D.shape)
print('XShape:', X.shape)
mat_dir = '../results/' + dataset_name + '_dic/'
if not os.path.exists(mat_dir):
os.makedirs(mat_dir)
mat_name = mat_dir + label + '.mat'
scipy.io.savemat(mat_name, {'dic': D, 'X': X})
print('MAT:', mat_name, 'saved!')
def ksvd(num_atoms):
return ApproximateKSVD(num_atoms)
mart_dict = getDict(center20)
SW_dict = getDict(center26)
pear_dict = getDict(center773)
PW_dict = getDict(center911)
# make an array for looping later
dictionaries = [spher_dict, PS_dict, net_dict, mart_dict, SW_dict, pear_dict, PW_dict]
## Test images
spher_test_img = img_array2[0:steph, 0:stepv]
patches = image.extract_patches_2d(spher_test_img, (16,16))
signals = patches.reshape(patches.shape[0], -1)
aksvd = ApproximateKSVD(n_components=64, max_iter = 100, transform_n_nonzero_coefs=4)
dictionary = aksvd.fit(signals[np.random.choice(range(signals.shape[0]), 1000, replace = True)]).components_
gamma = aksvd.transform(signals)
plt.imshow(image.reconstruct_from_patches_2d(
gamma.dot(dictionary).reshape((29200, 16, 16)), (161, 215)))
#generate test images and target vector at the same time
image_dict = []
targets = []
for i in range(3):
for j in range(3):
image_dict.append(img_array2[(i)*steph:(i+1)*steph, (j)*stepv:(j+1)*stepv])
targets.append(0)
image_dict.append(img_array4[(i)*steph:(i+1)*steph, (j)*stepv:(j+1)*stepv])
targets.append(1)