class LDAExtractor(AbstractSigExtractor):
def __init__(self, pca_file, lda_file):
super(LDAExtractor, self).__init__()
self.lda = LDA.load_from_file(lda_file)
self.pca = PCA()
self.pca.load_from_xml_file(pca_file)
def project_signature(self, feats):
pca_feats = self.pca.project(feats)
return self.lda.project(pca_feats, whiten=True)
def encode(self):
dla_dict = {}
dla_dict['extractor_type'] = 'lda_extractor'
dla_dict['lda'] = self.lda.encode()
dla_dict['pca'] = self.pca.encode()
return dla_dict
@classmethod
def decode(cls, lda_dict):
lda_dict['pca'] = PCA.decode(lda_dict['pca'])
lda_dict['lda'] = LDA.decode(lda_dict['lda'])
lda_extractor = cls.__new__(LDAExtractor)
lda_extractor.__dict__ = lda_dict
return lda_extractor
class DLAExtractor(AbstractSigExtractor):
def __init__(self, pca_file, dla_file):
"""
Args:
pca_file : the xml file that is saved as part of PCA learning
dla_file : pickle file storing the dla
"""
super(DLAExtractor, self).__init__()
self.dla = DLA.load_from_file(dla_file)
self.pca = PCA()
self.pca.load_from_xml_file(pca_file)
def project_signature(self, feats):
pca_feats = self.pca.project(feats)
return self.dla.project(pca_feats)
def encode(self):
dla_dict = {}
dla_dict['extractor_type'] = 'dla_extractor'
dla_dict['dla'] = self.dla.encode()
dla_dict['pca'] = self.pca.encode()
return dla_dict
@classmethod
def decode(cls, dla_dict):
dla_dict['pca'] = PCA.decode(dla_dict['pca'])
dla_dict['dla'] = DLA.decode(dla_dict['dla'])
dla_extractor = cls.__new__(DLAExtractor)
dla_extractor.__dict__ = dla_dict
return dla_extractor
class SubspaceExtractor(AbstractSigExtractor):
def __init__(self, model_dir):
super(SubspaceExtractor, self).__init__()
self.model_dir = model_dir
self.pca_file = os.path.join(self.model_dir, 'learned_pca.xml')
assert os.path.exists(self.pca_file), self.pca_file
self.pca = PCA()
self.pca.load_from_xml_file(self.pca_file)
self.subspace_file = os.path.join(self.model_dir, 'learned_subspace.xml')
assert os.path.exists(self.subspace_file), self.subspace_file
self.get_projection_matrix()
def get_projection_matrix(self):
WI = np.asarray(cv.Load(self.subspace_file, name='WI'))
WI_wh = np.asarray(cv.Load(self.subspace_file, name='WI_wh'))
WL = np.asarray(cv.Load(self.subspace_file, name='WL'))
self.projection_mat = np.dot(WL, np.dot(WI_wh, WI)).transpose()
def project(self, feats):
return np.dot(feats, self.projection_mat)
def project_signature(self, feats):
red_feats = self.pca.project(feats)
return self.project(red_feats)
class PCAExtractor(AbstractSigExtractor):
PCA_TAR_BALL_NAME = 'pca'
def __init__(self):
super(PCAExtractor, self).__init__()
pca_file = self.grab_projection_files()
self.pca = PCA()
self.pca.load_from_xml_file(pca_file)
def grab_projection_files(self):
download_dir = os.path.join(config.scratch_detector_path(), self.PCA_TAR_BALL_NAME)
bucket = config.s3_detector_bucket()
logger.info('Downloading files from s3')
s3client.download_tarball(bucket, self.PCA_TAR_BALL_NAME, download_dir)
pca_file = os.path.join(download_dir, 'pca.xml')
assert os.path.exists(pca_file), pca_file
return pca_file
def project_signature(self, feats):
return self.pca.project(feats)
def encode(self):
pca_dict = {}
pca_dict['extractor_type'] = 'pca_extractor'
pca_dict['pca'] = self.pca.encode()
return pca_dict
@classmethod
def decode(cls, pca_dict):
pca_dict['pca'] = PCA.decode(pca_dict['pca'])
pca_extractor = cls.__new__(cls)
pca_extractor.__dict__ = pca_dict
return pca_extractor
def __init__(self, pca_file, dla_file):
"""
Args:
pca_file : the xml file that is saved as part of PCA learning
dla_file : pickle file storing the dla
"""
super(DLAExtractor, self).__init__()
self.dla = DLA.load_from_file(dla_file)
self.pca = PCA()
self.pca.load_from_xml_file(pca_file)
def __init__(self, model_dir):
super(SubspaceExtractor, self).__init__()
self.model_dir = model_dir
self.pca_file = os.path.join(self.model_dir, 'learned_pca.xml')
assert os.path.exists(self.pca_file), self.pca_file
self.pca = PCA()
self.pca.load_from_xml_file(self.pca_file)
self.subspace_file = os.path.join(self.model_dir, 'learned_subspace.xml')
assert os.path.exists(self.subspace_file), self.subspace_file
self.get_projection_matrix()
def decode(cls, dla_dict):
dla_dict['pca'] = PCA.decode(dla_dict['pca'])
dla_dict['dla'] = DLA.decode(dla_dict['dla'])
dla_extractor = cls.__new__(DLAExtractor)
dla_extractor.__dict__ = dla_dict
return dla_extractor
def decode(cls, lda_dict):
lda_dict['pca'] = PCA.decode(lda_dict['pca'])
lda_dict['lda'] = LDA.decode(lda_dict['lda'])
lda_extractor = cls.__new__(LDAExtractor)
lda_extractor.__dict__ = lda_dict
return lda_extractor
def __init__(self, pca_file, lda_file):
super(LDAExtractor, self).__init__()
self.lda = LDA.load_from_file(lda_file)
self.pca = PCA()
self.pca.load_from_xml_file(pca_file)
def decode(cls, pca_dict):
pca_dict['pca'] = PCA.decode(pca_dict['pca'])
pca_extractor = cls.__new__(cls)
pca_extractor.__dict__ = pca_dict
return pca_extractor
def __init__(self):
super(PCAExtractor, self).__init__()
pca_file = self.grab_projection_files()
self.pca = PCA()
self.pca.load_from_xml_file(pca_file)