def dbProcess(self, password, process='s', model='s', host='localhost',
database='postgres', user='******', port=5432, dbase = 'imagedata_l'):
# Connect to the database
db = PostgresDB(password=password, host=host,
database=DATABASE_NAME, user=user, port=port)
conn = db.connect()
if process == 's':
self.dbSave(conn, model)
print('Data saved successfully to the Database!')
elif process == 'f':
recs = self.dbFetch(conn,dbase)
recs_flt = []
# Flatten the data structure and
for rec in recs:
recs_flt.append((rec[0],np.asarray(eval(rec[1]))))
# if model == 'm':
# print(recs)
# for rec in recs:
# recs_flt.append(np.asarray(eval(rec[1])))
# recs_flt.append((rec[0], [float(x) for y in rec[1] for x in y]))
# elif model == 's':
# for rec in recs:
# recs_flt.append((rec[0], [[float(x) for x in y] for y in rec[1]]))
# elif model == 'l' or model == 'h':
# for rec in recs:
# recs_flt.append((rec[0], [float(x) for x in rec[1]]))
return recs_flt
def dbProcess(self,
password,
process='s',
model='s',
host='localhost',
database='postgres',
user='******',
port=5432):
# Connect to the database
db = PostgresDB(password=password,
host=host,
database=database,
user=user,
port=port)
conn = db.connect()
if process == 's':
self.dbSave(conn, model)
print('Data saved successfully to the Database!')
elif process == 'f':
recs = self.dbFetch(conn, model)
recs_flt = []
# Flatten the data structure and
if model == 'm':
print(recs)
for rec in recs:
recs_flt.append(
(rec[0], [float(x) for y in rec[1] for x in y]))
elif model == 's':
for rec in recs:
recs_flt.append(
(rec[0], [[float(x) for x in y] for y in rec[1]]))
elif model == 'l' or model == 'h':
for rec in recs:
recs_flt.append((rec[0], [float(x) for x in rec[1]]))
return recs_flt
def get_model(args):
if args.model == 'fasttext':
model = FastTextModel(args.modelPath)
elif args.model == 'nle':
db = PostgresDB(util.read_db_config(args.db_cred))
model = NLEModel(args.modelPath, args.njobs, False, db)
model.load_indexes()
return model
def similarity(self, feature, technique, dbase, k, image, label=""):
db = PostgresDB(password="******", database="mwdb")
conn = db.connect()
if conn is None:
print("Can not connect to database")
exit()
# print(dbase)
cursor = conn.cursor()
cursor.execute("SELECT * FROM " + dbase)
data = cursor.fetchall()
image_id = [rec[0] for rec in data]
similarity = {}
if image in image_id:
image_index = image_id.index(image)
# print(image_index)
image_data = np.asarray(eval(data[image_index][1]))
else:
print("Not Same Label")
dbase = 'imagedata_' + feature
label = label.replace(" ", "_")
image_data = self.dbFetch(conn, dbase,
"WHERE imageid = '{0}'".format(image))
image_data = self.queryImageNotLabel(image_data, feature,
technique, label)
similarity[image] = self.euclidean_distance(image_data, image_data)
# print (image_id)
for i in range(len(image_id)):
image_cmp = np.asarray(eval(data[i][1]))
# print(data[i][0])
# print(data[i][1])
# if self.metrics:
# # similarity[row[0]] = 1- self.cosine_similarity(image, result)
# similarity[image_id[i]] = 1 - st.pearsonr(image,image_cmp)[0]
# # similarity[row[0]] = mean_squared_error(image,result)
# # similarity[row[0]] = 0 - self.psnr(image,result)
# else:
similarity[image_id[i]] = self.euclidean_distance(
image_data, image_cmp)
# print(similarity)
similarity = sorted(similarity.items(),
key=lambda x: x[1],
reverse=False)
# print(similarity)
self.dispImages(similarity, feature, technique, 11, k, label)
def run():
parser = argparse.ArgumentParser(description='Calcuclate embeddings')
parser.add_argument('input', help='Input containing osm samples')
parser.add_argument('output', help='File to safe embeddings in')
parser.add_argument('--model',
help='Embedding model to us',
default="NLE",
type=str)
parser.add_argument('--ftmodel',
help='Path to fasttext model',
type=str,
default="")
parser.add_argument('--db_cred',
help='Credentials for database',
type=str,
default="")
parser.add_argument('--njobs',
help='Number of threads to use',
default=1,
type=int)
args = parser.parse_args()
# create model
if args.model == "fasttext":
model = FastTextModel(args.ftmodel)
elif args.model == 'NLE':
db = PostgresDB(read_db_config(args.db_cred))
model = NLEModel(args.output, args.njobs, db)
else:
raise Exception("Model not found")
data = read_samples(args.input)
# train model
model.train(data)
# save model
model.save_model(args.output + "_" + args.model)
from dimReduction import dimReduction
from PostgresDB import PostgresDB
arg = input(
"Please enter the home directory for the images "
"(Default: C:\\Users\\pylak\\Documents\\Fall 2019\\MWDB\\Project\\Dataset\\) : "
)
if arg == '':
arg = 'C:\\Users\\pylak\\Documents\\Fall 2019\\MWDB\\Project\\Dataset\\'
dim = dimReduction(arg, '*.jpg')
db = PostgresDB(password='******',
host='localhost',
database='mwdb',
user='******',
port=5432)
conn = db.connect()
# Create for M and PCA
_, _, _, _ = dim.saveDim('m',
'pca',
'imagedata_m',
10,
password="******",
database="mwdb")
bin_matrix, feature_matrix = dim.binMat(conn, 'imagedata_m_pca')
# code to print
# print(bin_matrix[0][0])
dim.imgViz(bin_matrix)
for idx, sub in enumerate(feature_matrix):
print('\nLatent Semantic {x}'.format(x=idx + 1))
for s in sub:
print('Subject: ', s[0])
print('Weight: ', s[1])
def saveDim(self,
feature,
model,
dbase,
k,
password='******',
host='localhost',
database='mwdb',
user='******',
port=5432,
label=None,
meta=False,
negative_handle='n'):
imageDB = imageProcess(self.dirpath)
imgs = imageDB.dbProcess(password=password,
process='f',
model=feature,
dbase=dbase)
kmeans_model = 'kmeans_' + str(no_clusters)
technique_model = feature + '_' + model
if label is not None:
filteredImage = imageDB.CSV(label)
label = label.replace(" ", "_")
dbase += '_' + model + '_' + label
kmeans_model += '_' + label
technique_model += '_' + label
else:
dbase += '_' + model
# print(technique_model)
imgs_data = []
imgs_meta = []
i = -1
while i < len(imgs) - 1:
# print (x[1].shape)
i += 1
if label is not None and imgs[i][0] not in filteredImage:
# print("label")
del imgs[i]
i -= 1
continue
if feature != "s":
imgs_data.append(imgs[i][1].reshape((-1)))
else:
imgs_data.extend(imgs[i][1])
# print (image_cmp.shape)
imgs_meta.append(imgs[i][0])
# print(i)
# print(len(imgs))
#Handle Negative Value of NMF
if feature == 'm':
if negative_handle == 'h':
imgs_data = self.hist(imgs_data)
else:
imgs_data = self.normalize(imgs_data)
imgs_data = np.asarray(imgs_data)
# print(imgs_data.shape)
# print(imgs_meta)
# imgs_meta = [x[0] if x[0] in filteredImage for x in imgs]
imgs_zip = list(zip(imgs_meta, imgs_data))
db = PostgresDB(password=password,
host=host,
database=database,
user=user,
port=port)
conn = db.connect()
if meta:
imageDB.createInsertMeta(conn)
model = model.lower()
if feature == "s":
if imgs_data.shape[0] < no_clusters:
Kmeans = KMeans_SIFT(imgs_data.shape[0] // 2)
else:
Kmeans = KMeans_SIFT(no_clusters)
clusters = Kmeans.kmeans_process(imgs_data)
# print (imgs_zip)
imgs_data = Kmeans.newMatrixSift(imgs, clusters, kmeans_model)
imgs_zip = list(zip(imgs_meta, imgs_data))
if model == 'nmf':
w, h = self.nmf(imgs_data, k, technique_model)
imgs_red = np.dot(imgs_data, h.T).tolist()
print(np.asarray(w).shape)
print(np.asarray(h).shape)
imgs_sort = self.imgSort(w.T, imgs_meta)
feature_sort = self.imgFeatureSort(h, imgs_zip)
elif model == 'lda':
w, h = self.lda(imgs_data, k, technique_model)
imgs_red = np.dot(imgs_data, h.T).tolist()
print(np.asarray(w).shape)
print(np.asarray(h).shape)
imgs_sort = self.imgSort(w.T, imgs_meta)
feature_sort = self.imgFeatureSort(h, imgs_zip)
elif model == 'pca':
data, U, Vt = self.pca(imgs_data, k, technique_model)
imgs_red = data.tolist()
imgs_sort = self.imgSort(U.T, imgs_meta)
feature_sort = self.imgFeatureSort(Vt, imgs_zip)
elif model == 'svd':
# print(imgs_data.shape)
data, U, Vt = self.svd(imgs_data, k, technique_model)
imgs_red = data.tolist()
# print(im)
# U[:,:self.k].dot(Sigma[:self.k, :self.k]).dot(V[:self.k,:])
imgs_sort = self.imgSort(U.T, imgs_meta)
feature_sort = self.imgFeatureSort(Vt, imgs_zip)
# print("=======================")
# print(imgs_sort)
# print("=======================")
# print(feature_sort)
# Process the reduced Images
imgs_red = list(zip(imgs_meta, imgs_red))
# print (np.asarray(imgs_sort).shape)
# print(img_sort)
# print (np.asarray(feature_sort).shape)
# imgs_red = self.convString(imgs_red)
print(imgs_red)
self.createInsertDB(dbase, imgs_red, conn)
return imgs_sort, feature_sort
def saveDim(self,
feature,
model,
dbase,
k,
password='******',
host='localhost',
database='postgres',
user='******',
port=5432,
label=None,
meta=True,
negative_handle='n'):
imageDB = imageProcess(self.dirpath)
imgs = imageDB.dbProcess(password=password,
process='f',
model=feature,
dbase=dbase)
kmeans_model = 'kmeans_' + str(no_clusters) + '_' + feature
technique_model = feature + '_' + model
db = PostgresDB(password=password,
host=host,
database=database,
user=user,
port=port)
conn = db.connect()
if meta:
imageDB.createInsertMeta(conn)
if label is not None:
imgs = imageDB.CSV(conn, dbase, label)
label = label.replace(" ", "_")
dbase += '_' + model + '_' + label
kmeans_model += '_' + label
technique_model += '_' + label
else:
dbase += '_' + model
# print(technique_model)
imgs_data = []
imgs_meta = []
for img in imgs:
if feature == "s" or (feature == "m" and model in ("nmf", "lda")):
imgs_data.extend(img[1])
else:
imgs_data.append(img[1].reshape((-1)))
# print (image_cmp.shape)
imgs_meta.append(img[0])
# print(i)
# print(len(imgs))
# print(imgs_meta)
# print(len(imgs_meta))
#Handle Negative Value of NMF
# if feature == 'm' and (model == 'lda' or model == 'nmf'):
# print ("Normalize")
# if negative_handle == 'h':
# imgs_data = self.hist(imgs_data)
# else:
# imgs_data = self.normalize(imgs_data)
imgs_data = np.asarray(imgs_data)
# print(imgs_data.shape)
# print(imgs_data.shape)
# print(imgs_meta)
# imgs_meta = [x[0] if x[0] in filteredImage for x in imgs]
imgs_zip = list(zip(imgs_meta, imgs_data))
model = model.lower()
if feature == "s" or (feature == "m" and model in ("nmf", "lda")):
if imgs_data.shape[0] < no_clusters:
Kmeans = KMeans_SIFT(imgs_data.shape[0] // 2)
else:
Kmeans = KMeans_SIFT(no_clusters)
clusters = Kmeans.kmeans_process(imgs_data)
# print (imgs_zip)
imgs_data = Kmeans.newMatrixSift(imgs, clusters, kmeans_model)
imgs_zip = list(zip(imgs_meta, imgs_data))
if model == 'nmf':
w, h = self.nmf(imgs_data, k, technique_model)
imgs_red = np.dot(imgs_data, h.T).tolist()
# print(np.asarray(w).shape)
# print(np.asarray(h).shape)
imgs_sort = self.imgSort(w.T, imgs_meta)
feature_sort = self.imgFeatureSort(h, imgs_zip)
U = w
Vt = h
elif model == 'lda':
w, h = self.lda(imgs_data, k, technique_model)
imgs_red = np.dot(imgs_data, h.T).tolist()
# print(np.asarray(w).shape)
# print(np.asarray(h).shape)
imgs_sort = self.imgSort(w.T, imgs_meta)
feature_sort = self.imgFeatureSort(h, imgs_zip)
U = w
Vt = h
elif model == 'pca':
data, U, Vt = self.pca(imgs_data, k, technique_model)
imgs_red = data.tolist()
imgs_sort = self.imgSort(U.T, imgs_meta)
feature_sort = self.imgFeatureSort(Vt, imgs_zip)
elif model == 'svd':
# print(imgs_data.shape)
data, U, Vt = self.svd(imgs_data, k, technique_model)
imgs_red = data.tolist()
# print(im)
# U[:,:self.k].dot(Sigma[:self.k, :self.k]).dot(V[:self.k,:])
# print(U.T.shape)
# print(imgs_meta.shape)
imgs_sort = self.imgSort(U.T, imgs_meta)
feature_sort = self.imgFeatureSort(Vt, imgs_zip)
# print("=======================")
# print(imgs_sort)
# print("=======================")
# print(feature_sort)
# Process the reduced Images
imgs_red = list(zip(imgs_meta, imgs_red))
# print (np.asarray(imgs_sort).shape)
# print(img_sort)
# print (np.asarray(feature_sort).shape)
self.createInsertDB(dbase, imgs_red, conn)
return imgs_sort, feature_sort, U, Vt
from dimReduction import dimReduction
from PostgresDB import PostgresDB
arg = input(
"Please enter the home directory for the images "
"(Default: C:\\Users\\pylak\\Documents\\Fall 2019\\MWDB\\Project\\Phase1\\Hands_test2\\) : "
)
if arg == '':
arg = 'C:\\Users\\pylak\\Documents\\Fall 2019\\MWDB\\Project\\Phase1\\Hands_test2\\'
dim = dimReduction(arg, '*.jpg')
db = PostgresDB(password='******',
host='localhost',
database='postgres',
user='******',
port=5432)
conn = db.connect()
bin_matrix, feature_matrix = dim.binMat(conn, 'imagedata_m_pca')
# code to print
print(bin_matrix[0][0])
print(feature_matrix)
