def test_dot():
v1 = vector([1, 0])
v2 = vector([0, 1])
v3 = vector([-1, 0])
assert dot(v1, v1) == 1
assert dot(v1, v2) == 0
assert dot(v1, v3) == -1
def trainListFile(self, listTrainFile, listmanualfiles):
if len(listmanualfiles) != len(listTrainFile):
print("Co loi")
sys.exit()
self.reset()
queries = dlib.ranking_pairs()
for index in range(0, len(listTrainFile)):
self.reset()
data = dlib.ranking_pair()
inputNonRelevant = " ".join([line for line in open(listTrainFile[index], 'r').readlines()])
tpAllSent = myTokenizer(inputNonRelevant)
self.inputFromString(inputNonRelevant)
inputRelevant = " ".join([line for line in open(listmanualfiles[index], 'r').readlines()])
tpRelevant = myTokenizer(inputRelevant)
tpNonRelevant = list(set(tpAllSent).difference(set(tpRelevant)))
self.genAllVector()
for sent in tpRelevant:
data.relevant.append(dlib.vector(self.dicVector.get(sent.strip())))
for sent in tpNonRelevant:
data.nonrelevant.append(dlib.vector(self.dicVector.get(sent.strip())))
queries.append(data)
trainer = dlib.svm_rank_trainer()
trainer.c = 10
rank = trainer.train(queries)
_weight = []
for i in range(0, len(rank.weights)):
_weight.append(rank.weights[i])
return _weight
def generate_test_vectors():
vs = vectors()
vs.append(vector([0, 1, 2]))
vs.append(vector([3, 4, 5]))
vs.append(vector([6, 7, 8]))
assert len(vs) == 3
return vs
def train(tagged):
"""
Trains an SVM classifier based on the training data passed.
Mostly based on http://dlib.net/svm_binary_classifier.py.html.
:param tagged: list of TaggedFace to train on
:return: dlib.svm
"""
x = dlib.vectors() # will carry the facial encodings
y = dlib.array() # will carry the gender label
print("Preparing dataset...")
total = len(tagged)
for i, t in enumerate(tagged):
print(f"\rEncoding {t.path} ({i + 1}/{total})...", end="")
faces = encode(t.img)
x.append(dlib.vector(faces[0]))
y.append(t.tag)
img = t.img
for _ in range(5):
faces = encode(img)
if not faces:
break
x.append(dlib.vector(faces[0]))
y.append(t.tag)
img = cv2.resize(img, None, fx=0.7, fy=0.7)
print("Training SVM...")
trainer = dlib.svm_c_trainer_radial_basis()
#trainer.be_verbose()
trainer.set_c(10)
model = trainer.train(x, y)
with open(PATH_SVMFILE, "wb") as filehandle:
pickle.dump(model, filehandle)
return None
def sentence_to_vectors(sentence):
# Create an empty array of vectors
vects = dlib.vectors()
for word in sentence.split():
# Our vectors are very simple 1-dimensional vectors. The value of the single
# feature is 1 if the first letter of the word is capitalized and 0 otherwise.
if (word[0].isupper()):
vects.append(dlib.vector([1]))
else:
vects.append(dlib.vector([0]))
return vects
def sentence_to_vectors(sentence):
# Create an empty array of vectors
vects = dlib.vectors()
for word in sentence.split():
# Our vectors are very simple 1-dimensional vectors. The value of the single
# feature is 1 if the first letter of the word is capitalized and 0 otherwise.
if (word[0].isupper()):
vects.append(dlib.vector([1]))
else:
vects.append(dlib.vector([0]))
return vects
def dlibVectorFormating(data, tolist=True, key_descr='descr'):
for d in data:
if tolist:
d[key_descr] = [list(dd) for dd in d[key_descr]]
else:
d[key_descr] = [dlib.vector(dd) for dd in d[key_descr]]
return data
def CLUSTER_TRACKS(DT, threshold):
track_feats = []
for i in DT.keys():
track_feats.append(dlib.vector(DT[i]['BBOX_FEAT'].mean(0)))
CL = defaultdict(dict)
cluster_ids = dlib.chinese_whispers_clustering(track_feats, threshold)
for i in cluster_ids:
try:
CL[i]['BBOX'] = CL[i]['BBOX'] + DT[list(DT.keys())[i]]['BBOX']
CL[i]['Frame_ID'] = CL[i]['Frame_ID'] + DT[list(DT.keys())[i]]['Frame_ID']
CL[i]['BBOX_FEAT'] = CL[i]['BBOX_FEAT'] + DT[list(DT.keys())[i]]['BBOX_FEAT']
CL[i]['ANGLE'] = CL[i]['ANGLE'] + DT[list(DT.keys())[i]]['ANGLE']
CL[i]['IMG'] = CL[i]['IMG'] + DT[list(DT.keys())[i]]['IMG']
CL[i]['AVG_SIZE'] = DT[list(DT.keys())[i]]['AVG_SIZE']
CL[i]['AREA'] = DT[list(DT.keys())[i]]['AREA']
CL[i]['LEN'] = DT[list(DT.keys())[i]]['LEN'] + DT[list(DT.keys())[i]]['LEN']
except:
CL[i]['BBOX'] = DT[list(DT.keys())[i]]['BBOX']
CL[i]['Frame_ID'] = DT[list(DT.keys())[i]]['Frame_ID']
CL[i]['BBOX_FEAT'] = DT[list(DT.keys())[i]]['BBOX_FEAT']
CL[i]['ANGLE'] = DT[list(DT.keys())[i]]['ANGLE']
CL[i]['IMG'] = DT[list(DT.keys())[i]]['IMG']
CL[i]['AVG_SIZE'] = DT[list(DT.keys())[i]]['AVG_SIZE']
CL[i]['AREA'] = DT[list(DT.keys())[i]]['AREA']
CL[i]['LEN'] = DT[list(DT.keys())[i]]['LEN']
return CL
def preprocess_faces(self, faces):
# Cluster the faces with chinese whispers
encodings = [dlib.vector(face['encoding']) for face in faces]
labels = dlib.chinese_whispers_clustering(encodings, 0.5)
selected_faces = []
# Select face most close to average group
groups = list(set(labels))
for group in groups:
# Get indices for each group
indices = [i for i in range(len(labels)) if labels[i] == group]
group_encodings = [faces[i]['encoding'] for i in indices]
# Get centroid for group encodings
avg_group_encoding = np.average(group_encodings, axis=0)
# Get the closest face to the centroid
avg_distance = face_recognition.face_distance(
group_encodings, avg_group_encoding)
min_index = np.argmin(avg_distance)
face_index = indices[min_index]
selected_faces.append(faces[face_index])
return selected_faces
def clustring(self, faces_info):
for data in faces_info:
encode = data['face_encoding']
self.face_encodings.append(dlib.vector(encode))
labels = dlib.chinese_whispers_clustering(self.face_encodings, 0.5)
labels = np.array(labels)
print("All cluster labels :", labels)
unique_labels = np.unique(labels)
print("Number of unique faces found : ", len(unique_labels))
print("Saving faces..........")
for label in unique_labels:
index = np.where(labels == label)[0]
for i in index:
image_path = self.faces_info[i]['img_path']
image_name = image_path.split('/')[-1].split('.')[0]
image_ext = image_path.split('/')[-1].split('.')[1]
image = cv2.imread(image_path)
output_dir = os.getcwd() + '/' + str(label)
if not os.path.isdir(output_dir):
os.mkdir(str(label))
cv2.imwrite(output_dir + '/' + image_name + '.' + image_ext,
image)
def predict_gender(self, encoding, thresh=0.4):
result = self.classifier(dlib.vector(encoding))
if result > thresh:
return "male"
if result < -thresh:
return "female"
return "unknown"
def chinese_whispers(encodings, threshold=0.5):
"""
Chinese Whispers - an Efficient Graph Clustering Algorithm
and its Application to Natural Language Processing Problems
"""
encodings = [dlib.vector(enc) for enc in encodings]
return dlib.chinese_whispers_clustering(encodings, threshold)
def __clusterize(self, files_faces, debug_out_folder=None):
self.__start_stage(len(files_faces))
encs = []
indexes = list(range(len(files_faces)))
random.shuffle(indexes)
for i in indexes:
for j in range(len(files_faces[i]['faces'])):
encs.append(dlib.vector(
files_faces[i]['faces'][j]['encoding']))
labels = dlib.chinese_whispers_clustering(
encs, self.__threshold_clusterize)
labels = self.__reassign_by_count(labels)
lnum = 0
for i in indexes:
if self.__step_stage():
break
for j in range(len(files_faces[i]['faces'])):
files_faces[i]['faces'][j]['name'] = \
'unknown_{:05d}'.format(labels[lnum])
lnum += 1
if debug_out_folder:
filename = files_faces[i]['filename']
media = tools.load_media(filename,
self.__max_size,
self.__max_video_frames,
self.__video_frames_step)
debug_out_file_name = self.__extract_filename(filename)
self.__save_debug_images(
files_faces[i]['faces'], media,
debug_out_folder, debug_out_file_name)
self.__end_stage()
def make_psi(self, x, label):
"""Compute PSI(x,label)."""
psi = dlib.vector()
# Set it to have 9 dimensions. Note that the elements of the vector
# are 0 initialized.
psi.resize(self.num_dimensions)
# first
label_num = label[0]
# psi[:label_num * 128] = label_num * 128 * [0]
for index in range(128):
psi[label_num * 128 + index] = x[0][index]
# psi[label_num*128, (label_num+1)*128] = x[0].tolist()
# psi[(label_num+1)*128:] = (26 - label_num) * 128 * [0]
label_num = label[1]
for index in range(128):
psi[label_num * 128 + 128 * 27 + index] = x[1][index]
# get changing label
if label[0] != label[1]:
psi[-1] = 1
else:
psi[-1] = 0
return psi
def predict_gender(encoding):
result = _classifier(dlib.vector(encoding))
if result > 0.5:
return "male"
if result < -0.5:
return "female"
return "unknown"
def predict_gender(encoding, threshold=0.5):
result = _classifier(dlib.vector(encoding))
if result > threshold:
return "male"
if result < -threshold:
return "female"
return "unknown"
def calc_embded2(file_list):
embd_list = []
file_name = []
for f in file_list:
img = cv2.imread(f)
ret = face_recognition.face_encodings(img)
if len(ret) == 0:
continue
file_name.append(f)
embd_list.append(dlib.vector(ret[0]))
return file_name, embd_list
def test_vector_set_size():
v = vector(3)
v.set_size(0)
assert len(v) == 0
assert v.shape == (0, 1)
v.resize(10)
assert len(v) == 10
assert v.shape == (10, 1)
for i in range(10):
assert v[i] == 0
def __recognize(self):
"""
Recognize face and return it's descriptor
:return:
"""
try:
face_roi: RoiData = self.__frames.get()
if not self.__config.recognize_faces:
return
img: np.ndarray = face_roi.img
b, g, r = cv2.split(img)
img_rgb = cv2.merge((r, g, b))
# win = dlib.image_window()
# win.clear_overlay()
# win.set_image(img_rgb)
# win.add_overlay(face_roi.shape)
# win.wait_until_closed()
face_desc = self.__face_rec_model.compute_face_descriptor(img_rgb, face_roi.shape)
faces = self.__db_worker.select_all_faces()
wrong_face = True
for face in faces:
desc: str = face[3]
values = [float(x) for x in desc.split('\n')]
vector = dlib.vector(values)
faces_dist = distance.euclidean(face_desc, vector)
if faces_dist < 0.6:
wrong_face = False
break
if wrong_face:
if len(self.__faces) < 5:
self.__faces.append(img)
if not self.__thread_started:
self.__thread_started = True
self.__thread = Thread(target=self.__send_notification)
self.__thread.name = "NotificationThread"
self.__thread.start()
except Exception as ex:
self.__recognition_error(f"{ex}")
def estimate_gender(face):
"""
Estimates a characteristic based on the face that is passed.
:param face: dlibs 128-long face encoding
:return: float, estimated gender. The gender model has been trained as
value 1 for females, and -1 for males. So, a value of -0.5 means "mainly
male" and can be considered as such. Values between -0.3 and 0.3 mean
the model is not certain enough, and should be considered as "unknown"
or "uncertain"
"""
vector = dlib.vector(face)
return gender_model(vector)
def compute_similarities(data_dir, similarity_threshold=0.6, identity_threshold=0.4, criminal_fraction=0.1, **kwargs):
t = Timer()
all_descriptors = db.get_all_descriptors()
descriptors = [json.loads(f[1]) for f in all_descriptors]
face_ids = [f[0] for f in all_descriptors]
num_faces = len(all_descriptors)
#print("get_all_descriptors():", t)
#print("Faces: %d" % len(all_descriptors), end='')
if num_faces < 2:
#print()
return num_faces, 0, 0
X = Y = np.array(descriptors)
#print("convert to array:", t)
X2 = Y2 = np.sum(np.square(X), axis=-1)
dists = np.sqrt(np.maximum(X2[:, np.newaxis] + Y2[np.newaxis] - 2 * np.dot(X, Y.T), 0))
#print("calculate dists:", t)
db.delete_similarities()
#print("delete similarities:", t)
num_similarities = 0
for i, j in zip(*np.where(dists < float(similarity_threshold))):
if i != j:
db.insert_similarity([face_ids[i], face_ids[j], dists[i, j]])
num_similarities += 1
#print("save similarities:", t)
# cluster faces and update labels
descriptors_dlib = [dlib.vector(d) for d in descriptors]
clusters = dlib.chinese_whispers_clustering(descriptors_dlib, float(identity_threshold))
db.update_labels(zip(clusters, face_ids))
num_clusters = len(set(clusters))
if args.save_clusters:
for cluster_num, face_id in zip(clusters, face_ids):
facefile = os.path.realpath(os.path.join(data_dir, args.save_faces, "face_%05d.jpg" % face_id))
clusterdir = os.path.join(data_dir, args.save_clusters, str(cluster_num))
makedirs(clusterdir)
os.symlink(facefile, os.path.join(clusterdir, 'tmpfile'))
os.rename(os.path.join(clusterdir, 'tmpfile'), os.path.join(clusterdir, "face_%05d.jpg" % face_id))
# remove clusters with more than given amount of criminals
criminal_clusters = db.get_clusters_with_criminals(criminal_fraction)
for cluster in criminal_clusters:
db.remove_cluster(cluster['cluster_num'])
db.commit()
#print("commit:", t)
#print(", Similarities: %d, Time: %.2fs" % (num_similarities, t.total()))
return num_faces, num_similarities, num_clusters
def match(candidate):
bestThresh = 9999
bestIndex = -1
if (len(helpers.unique_persons) > 0):
for index, person in enumerate(helpers.unique_persons):
currThresh = helpers.euclidean_dist(candidate,
dlib.vector(person["Mean"]))
if (currThresh < helpers.MAX_MATCHING_THRESH):
if (currThresh < bestThresh):
bestIndex = index
bestThresh = currThresh
return bestIndex
def cluster_faces(src_dir):
# Load face metadata
faces_df = pd.read_csv(os.path.join(src_dir, 'metadata.csv'))
# Check if clustering already exists
if 'cluster' not in faces_df.columns:
# Chinese whispers clustering
faces_df['embedding'] = faces_df['json_embedding'].apply(json.loads)
X = np.array([x for x in faces_df['embedding']])
faces_df['cluster'] = dlib.chinese_whispers_clustering(
[dlib.vector(x) for x in X], 0.5)
# Persist clustering
faces_df.to_csv(os.path.join(src_dir, 'metadata.csv'), index=False)
def training_data():
r = Random(0)
predictors = vectors()
sparse_predictors = sparse_vectors()
response = array()
for i in range(30):
for c in [-1, 1]:
response.append(c)
values = [r.random() + c * 0.5 for _ in range(3)]
predictors.append(vector(values))
sp = sparse_vector()
for i, v in enumerate(values):
sp.append(pair(i, v))
sparse_predictors.append(sp)
return predictors, sparse_predictors, response
def training_data():
r = Random(0)
predictors = vectors()
sparse_predictors = sparse_vectors()
response = array()
for i in range(30):
for c in [-1, 1]:
response.append(c)
values = [r.random() + c * 0.5 for _ in range(3)]
predictors.append(vector(values))
sp = sparse_vector()
for i, v in enumerate(values):
sp.append(pair(i, v))
sparse_predictors.append(sp)
return predictors, sparse_predictors, response
def test_vector_slice():
v = vector([1, 2, 3, 4, 5])
v_slice = v[1:4]
assert len(v_slice) == 3
for idx, val in enumerate([2, 3, 4]):
assert v_slice[idx] == val
v_slice = v[-3:-1]
assert len(v_slice) == 2
for idx, val in enumerate([3, 4]):
assert v_slice[idx] == val
v_slice = v[1:-2]
assert len(v_slice) == 2
for idx, val in enumerate([2, 3]):
assert v_slice[idx] == val
def make_psi(self, x, label):
"""Compute PSI(x,label)."""
# All we are doing here is taking x, which is a 3 dimensional sample
# vector in this example program, and putting it into one of 3 places in
# a 9 dimensional PSI vector, which we then return. So this function
# returns PSI(x,label). To see why we setup PSI like this, recall how
# predict_label() works. It takes in a 9 dimensional weight vector and
# breaks the vector into 3 pieces. Each piece then defines a different
# classifier and we use them in a one-vs-all manner to predict the
# label. So now that we are in the structural SVM code we have to
# define the PSI vector to correspond to this usage. That is, we need
# to setup PSI so that argmax_y dot(weights,PSI(x,y)) ==
# predict_label(weights,x). This is how we tell the structural SVM
# solver what kind of problem we are trying to solve.
#
# It's worth emphasizing that the single biggest step in using a
# structural SVM is deciding how you want to represent PSI(x,label). It
# is always a vector, but deciding what to put into it to solve your
# problem is often not a trivial task. Part of the difficulty is that
# you need an efficient method for finding the label that makes
# dot(w,PSI(x,label)) the biggest. Sometimes this is easy, but often
# finding the max scoring label turns into a difficult combinatorial
# optimization problem. So you need to pick a PSI that doesn't make the
# label maximization step intractable but also still well models your
# problem.
#
# Create a dense vector object (note that you can also use unsorted
# sparse vectors (i.e. dlib.sparse_vector objects) to represent your
# PSI vector. This is useful if you have very high dimensional PSI
# vectors that are mostly zeros. In the context of this example, you
# would simply return a dlib.sparse_vector at the end of make_psi() and
# the rest of the example would still work properly. ).
psi = dlib.vector()
# Set it to have 9 dimensions. Note that the elements of the vector
# are 0 initialized.
psi.resize(self.num_dimensions)
dims = len(x)
if label == 0:
for i in range(0, dims):
psi[i] = x[i]
elif label == 1:
for i in range(dims, 2 * dims):
psi[i] = x[i - dims]
else: # the label must be 2
for i in range(2 * dims, 3 * dims):
psi[i] = x[i - 2 * dims]
return psi
def make_psi(self, x, label):
"""Compute PSI(x,label)."""
# All we are doing here is taking x, which is a 3 dimensional sample
# vector in this example program, and putting it into one of 3 places in
# a 9 dimensional PSI vector, which we then return. So this function
# returns PSI(x,label). To see why we setup PSI like this, recall how
# predict_label() works. It takes in a 9 dimensional weight vector and
# breaks the vector into 3 pieces. Each piece then defines a different
# classifier and we use them in a one-vs-all manner to predict the
# label. So now that we are in the structural SVM code we have to
# define the PSI vector to correspond to this usage. That is, we need
# to setup PSI so that argmax_y dot(weights,PSI(x,y)) ==
# predict_label(weights,x). This is how we tell the structural SVM
# solver what kind of problem we are trying to solve.
#
# It's worth emphasizing that the single biggest step in using a
# structural SVM is deciding how you want to represent PSI(x,label). It
# is always a vector, but deciding what to put into it to solve your
# problem is often not a trivial task. Part of the difficulty is that
# you need an efficient method for finding the label that makes
# dot(w,PSI(x,label)) the biggest. Sometimes this is easy, but often
# finding the max scoring label turns into a difficult combinatorial
# optimization problem. So you need to pick a PSI that doesn't make the
# label maximization step intractable but also still well models your
# problem.
#
# Create a dense vector object (note that you can also use unsorted
# sparse vectors (i.e. dlib.sparse_vector objects) to represent your
# PSI vector. This is useful if you have very high dimensional PSI
# vectors that are mostly zeros. In the context of this example, you
# would simply return a dlib.sparse_vector at the end of make_psi() and
# the rest of the example would still work properly. ).
psi = dlib.vector()
# Set it to have 9 dimensions. Note that the elements of the vector
# are 0 initialized.
psi.resize(self.num_dimensions)
dims = len(x)
if label == 0:
for i in range(0, dims):
psi[i] = x[i]
elif label == 1:
for i in range(dims, 2 * dims):
psi[i] = x[i - dims]
else: # the label must be 2
for i in range(2 * dims, 3 * dims):
psi[i] = x[i - 2 * dims]
return psi
def cluster_embeddings(encodings_path=None):
# Load previously generated embeddings
print("Loading encodings...")
data = pickle.loads(open(Path(encodings_path), "rb").read())
data = np.array(data)
# Specifically grab the encodings from the data array
# If using dlib's Chinese Whispers Clustering, convert to dlib vector format
encodings = [dlib.vector(d["encoding"].squeeze()) for d in data]
# If using KNN, keep in Numpy format
# encodings = [d["encoding"] for d in data]
# encodings = np.asarray(encodings).squeeze()
# Calculate a threshold value for Chinese Whispers
neigh = NearestNeighbors(n_neighbors=5)
nbrs = neigh.fit(encodings)
distances, indices = nbrs.kneighbors(encodings)
distances = np.sort(distances, axis=0)
distances = distances[:, 2]
mean_distance = np.mean(distances)
# plt.plot(distances)
# plt.show()
# Clustering with Chinese Whispers algorithm
labels = dlib.chinese_whispers_clustering(encodings, mean_distance)
# kmeans = KMeans(n_clusters=5, random_state=0).fit(encodings)
# label_ids = np.unique(kmeans.labels_)
# labels = kmeans.labels_
# Determine the total number of unique faces, as well
# as their occurrences
label_ids, counts = np.unique(labels, return_counts=True)
num_unique_faces = len(label_ids)
# Split images into clusters based on labels
image_paths = [d["image_path"] for d in data]
output_folder = image_paths[0].parent.parent.joinpath("clustered_faces")
Path(output_folder).mkdir(parents=True, exist_ok=True)
for i in range(len(image_paths)):
current_label = labels[i]
current_file = image_paths[i]
new_path = output_folder.joinpath(
str(current_label) + "_" + current_file.name)
shutil.copy(current_file, new_path)
def cluster():
s = time.time()
query = ''
descriptors = []
dvec = dlib.vectors()
date = input("enter a date in dd-mm-yyy format")
from_time = input("enter start time in hh:mm format")
to_time = input("enter end time in hh:mm format")
data = ptf.retrive(date, from_time, to_time)
for d in data:
descriptors.append(dlib.vector(d))
# Cluster the faces.
labels = dlib.chinese_whispers_clustering(descriptors, 0.5)
e = time.time()
print(labels)
print(len(descriptors))
print(len(labels))
labset = set(labels)
print(labset)
num_classes = len(set(labels)) #total number of clusters
print("Number of clusters: {}".format(num_classes))
print(e - s)
return num_classes
def __init__(self, data_path):
names = ['time', 'track']
for i in range(128):
names += ['d{0}'.format(i)]
#
self.data = read_table(data_path, delim_whitespace=True,
header=None, names=names)
self.data.sort_values(by=['track', 'time'], inplace=True)
# create a descriptor list with dlibs descriptor vector
descriptors = []
embeddings = self.data.iloc[:, 2:].values
for each_i in embeddings:
face_descriptor = dlib.vector(each_i)
descriptors.append(face_descriptor)
# returns series of labels [0 0 2 2 2] for each row of embeddings
labels = dlib.chinese_whispers_clustering(descriptors, 0.5)
# put the series into a column
self.data['cluster'] = pandas.Series(labels, index=self.data.index)
# TODO: this can be improved by taking highest count of label in each track
# get the label for each track
track_label = self.data.groupby(by='track', as_index=False).first()[
['track', 'cluster']].values
# get unique labels
self.labels = np.unique(track_label[:][:, [1]])
self.starting_point = Annotation(modality='face')
for track, segment in self.data.groupby('track').apply(_to_segment).iteritems():
if not segment:
continue
self.starting_point[segment, track] = track_label[track][1]
def train(self, directoryPlain, directoryManual):
self.reset()
listFile = []
listPlainFile = listAllFileInFolder(directoryPlain)
listManualFile = listAllFileInFolder(directoryManual)
dicPlainFile = {}
dicManualFile = {}
for file in listPlainFile:
fname = file.strip().split('/')[-1]
listFile.append(fname)
dicPlainFile[fname] = file
for file in listManualFile:
fname = file.strip().split('/')[-1]
listFile.append(fname)
dicManualFile[fname] = file
listFile = list(set(listFile))
queries = dlib.ranking_pairs()
countt = 0
outfile = open("completefile.txt", 'w')
for file in listFile:
outvecfile = open("/home/hien/Data/Work/Wordnet_naiscorp/test/valuevector/"+file.strip().split('/')[-1], 'w')
countt = countt + 1
outfile.write(file+'\n')
print (file, countt)
self.reset()
data = dlib.ranking_pair()
inputNonRelevant = " ".join([line for line in open(dicPlainFile.get(file), 'r').readlines()])
tpAllSent = myTokenizer(inputNonRelevant)
self.inputFromString(inputNonRelevant)
inputRelevant = " ".join([line for line in open(dicManualFile.get(file), 'r').readlines()])
tpRelevant = myTokenizer(inputRelevant)
tpNonRelevant = list(set(tpAllSent).difference(set(tpRelevant)))
self.genAllVector()
for sent in tpAllSent:
outvecfile.write(str(self.dicVector.get(sent.strip()))+"\t"+sent.strip()+'\n')
outvecfile.close()
for sent in tpRelevant:
# print (sent)
# print(self.dicVector.get(sent))
# print(type(self.dicVector.get(sent)))
data.relevant.append(dlib.vector(self.dicVector.get(sent.strip())))
# outvecfile.write(str(self.dicVector.get(sent.strip()))+"\t"+sent.strip()+'\n')
# outvecfile.
for sent in tpNonRelevant:
# print(self.dicVector.get(sent))
data.nonrelevant.append(dlib.vector(self.dicVector.get(sent.strip())))
queries.append(data)
trainer = dlib.svm_rank_trainer()
trainer.c = 10
rank = trainer.train(queries)
_weight = []
for i in range(0, len(rank.weights)):
_weight.append(rank.weights[i])
# print(type(rank.weights))
# print (rank.weights[0])
# print (rank.weights)
# print(_weight)
# return rank.weights
return _weight
import settings
train_data = pd.read_csv('training_features.csv', index_col=0, encoding="ISO-8859-1")
query_id_train = train_data["query_id"].tolist()
doc_id_train = train_data["doc_id"].tolist()
train_features = train_data[settings.feature_selected]
# train_true = list(train_data["label"])
train_true = train_data["label"].tolist()
# testing
test_data = pd.read_csv('test_features.csv', index_col=0, encoding="ISO-8859-1")
query_id_test = test_data["query_id"].tolist()
doc_id_test = test_data["doc_id"].tolist()
test_features = test_data[settings.feature_selected]
test_true = test_data["label"]
data = dlib.ranking_pair()
for i in range(len(train_true)):
if train_true[i] == 1:
data.relevant.append(dlib.vector(train_features[i]))
elif train_true[i] == 0:
data.nonrelevant.append(dlib.vector(train_features[i]))
trainer = dlib.svm_rank_trainer()
trainer.c = 10
rank = trainer.train(data)
print("Ranking score for a relevant vector: {}".format(
rank(data.relevant[0])))
print("Ranking score for a non-relevant vector: {}".format(
rank(data.nonrelevant[0])))
def test_vector_getitem():
v = vector([1, 2, 3])
assert v[0] == 1
assert v[-1] == 3
assert v[1] == v[-2]
# run compile_dlib_python_module.bat. This should work on any operating system # so long as you have CMake and boost-python installed. On Ubuntu, this can be # done easily by running the command: sudo apt-get install libboost-python-dev cmake import dlib # Now let's make some testing data. To make it really simple, let's suppose that # we are ranking 2D vectors and that vectors with positive values in the first # dimension should rank higher than other vectors. So what we do is make # examples of relevant (i.e. high ranking) and non-relevant (i.e. low ranking) # vectors and store them into a ranking_pair object like so: data = dlib.ranking_pair() # Here we add two examples. In real applications, you would want lots of # examples of relevant and non-relevant vectors. data.relevant.append(dlib.vector([1, 0])) data.nonrelevant.append(dlib.vector([0, 1])) # Now that we have some data, we can use a machine learning method to learn a # function that will give high scores to the relevant vectors and low scores to # the non-relevant vectors. trainer = dlib.svm_rank_trainer() # Note that the trainer object has some parameters that control how it behaves. # For example, since this is the SVM-Rank algorithm it has a C parameter that # controls the trade-off between trying to fit the training data exactly or # selecting a "simpler" solution which might generalize better. trainer.c = 10 # So let's do the training. rank = trainer.train(data)
def test_vector_serialization():
v = vector([1, 2, 3])
ser = pickle.dumps(v, 2)
deser = pickle.loads(ser)
assert str(v) == str(deser)
# sudo apt-get install libboost-python-dev cmake
#
import dlib
try:
import cPickle as pickle
except ImportError:
import pickle
x = dlib.vectors()
y = dlib.array()
# Make a training dataset. Here we have just two training examples. Normally
# you would use a much larger training dataset, but for the purpose of example
# this is plenty. For binary classification, the y labels should all be either +1 or -1.
x.append(dlib.vector([1, 2, 3, -1, -2, -3]))
y.append(+1)
x.append(dlib.vector([-1, -2, -3, 1, 2, 3]))
y.append(-1)
# Now make a training object. This object is responsible for turning a
# training dataset into a prediction model. This one here is a SVM trainer
# that uses a linear kernel. If you wanted to use a RBF kernel or histogram
# intersection kernel you could change it to one of these lines:
# svm = dlib.svm_c_trainer_histogram_intersection()
# svm = dlib.svm_c_trainer_radial_basis()
svm = dlib.svm_c_trainer_linear()
svm.be_verbose()
svm.set_c(10)
def test_vector_empty_init():
v = vector()
assert len(v) == 0
assert v.shape == (0, 1)
assert str(v) == ""
assert repr(v) == "dlib.vector([])"
def test_vector_init_with_negative_number():
with raises(Exception):
vector(-3)
def test_vector_invalid_getitem():
v = vector([1, 2, 3])
with raises(IndexError):
v[-4]
with raises(IndexError):
v[3]
def test_vector_init_with_number():
v = vector(3)
assert len(v) == 3
assert v.shape == (3, 1)
assert str(v) == "0\n0\n0"
assert repr(v) == "dlib.vector([0, 0, 0])"
# command: # sudo apt-get install cmake # import dlib # Now let's make some testing data. To make it really simple, let's suppose # that we are ranking 2D vectors and that vectors with positive values in the # first dimension should rank higher than other vectors. So what we do is make # examples of relevant (i.e. high ranking) and non-relevant (i.e. low ranking) # vectors and store them into a ranking_pair object like so: data = dlib.ranking_pair() # Here we add two examples. In real applications, you would want lots of # examples of relevant and non-relevant vectors. data.relevant.append(dlib.vector([1, 0])) data.nonrelevant.append(dlib.vector([0, 1])) # Now that we have some data, we can use a machine learning method to learn a # function that will give high scores to the relevant vectors and low scores to # the non-relevant vectors. trainer = dlib.svm_rank_trainer() # Note that the trainer object has some parameters that control how it behaves. # For example, since this is the SVM-Rank algorithm it has a C parameter that # controls the trade-off between trying to fit the training data exactly or # selecting a "simpler" solution which might generalize better. trainer.c = 10 # So let's do the training. rank = trainer.train(data)
# sudo apt-get install cmake
#
import dlib
try:
import cPickle as pickle
except ImportError:
import pickle
x = dlib.vectors()
y = dlib.array()
# Make a training dataset. Here we have just two training examples. Normally
# you would use a much larger training dataset, but for the purpose of example
# this is plenty. For binary classification, the y labels should all be either +1 or -1.
x.append(dlib.vector([1, 2, 3, -1, -2, -3]))
y.append(+1)
x.append(dlib.vector([-1, -2, -3, 1, 2, 3]))
y.append(-1)
# Now make a training object. This object is responsible for turning a
# training dataset into a prediction model. This one here is a SVM trainer
# that uses a linear kernel. If you wanted to use a RBF kernel or histogram
# intersection kernel you could change it to one of these lines:
# svm = dlib.svm_c_trainer_histogram_intersection()
# svm = dlib.svm_c_trainer_radial_basis()
svm = dlib.svm_c_trainer_linear()
svm.be_verbose()
svm.set_c(10)
def test_vectors_extend():
vs = vectors()
vs.extend([vector([1, 2, 3]), vector([4, 5, 6])])
assert len(vs) == 2
descriptors = []
images = []
# Now find all the persons 1024D descriptors.
personDesc = open(descriptor_file_path, "r")
for line in personDesc:
descriptorElements = line.split("|")
print("Processing image: {}".format(descriptorElements[0]))
# Compute the 128D vector that describes the face in img identified by
# shape.
descriptor = np.array(descriptorElements[1:])
descriptor = descriptor.astype(np.float)
descriptors.append(dlib.vector(descriptor))
images.append(descriptorElements[0])
#descriptors = dlib.vector(descriptors)
# Now let's cluster the faces.
labels = dlib.chinese_whispers_clustering(descriptors, 0.20)
num_classes = len(set(labels))
print("Number of clusters: {}".format(num_classes))
# Find biggest class
biggest_class = None
biggest_class_length = 0
for i in range(0, num_classes):
class_length = len([label for label in labels if label == i])
if class_length > biggest_class_length:
def test_vector_init_with_list():
v = vector([1, 2, 3])
assert len(v) == 3
assert v.shape == (3, 1)
assert str(v) == "1\n2\n3"
assert repr(v) == "dlib.vector([1, 2, 3])"