def __init__(self,
create_anomaly_model_func,
patches,
cell_size=0.2,
fake=False):
""" Create a new spatial bin anomaly model
Args:
create_anomaly_model_func (func): Method that returns an anomaly model
patches (PatchArray): The patches are needed to get the rasterization
cell_size (float): Width and height of spatial bin in meter
"""
AnomalyModelBase.__init__(self)
self.CELL_SIZE = cell_size
self.KEY = "%.2f" % self.CELL_SIZE
if fake: self.KEY = "fake_" + self.KEY
self.CREATE_ANOMALY_MODEL_FUNC = create_anomaly_model_func
self.FAKE = fake
# Get extent
x_min, y_min, x_max, y_max = patches.get_extent(cell_size, fake=fake)
# Create the bins
bins_y = np.arange(y_min, y_max, cell_size)
bins_x = np.arange(x_min, x_max, cell_size)
shape = (len(bins_y), len(bins_x))
# Create the grid
raster = np.zeros(shape, dtype=object)
for v, y in enumerate(bins_y):
for u, x in enumerate(bins_x):
b = box(
x, y, x + cell_size, y +
cell_size) #Point(x + cell_size / 2, y + cell_size / 2)#
b.u = u
b.v = v
b.patches = list()
raster[v, u] = b
# Create a search tree of spatial boxes
self._grid = STRtree(raster.ravel().tolist())
m = create_anomaly_model_func()
self.NAME = "SpatialBin/%s/%s" % (m.__class__.__name__.replace(
"AnomalyModel", ""), self.KEY)
def __init__(self, create_anomaly_model_func, cell_size=0.2, fake=False):
""" Create a new spatial bin anomaly model
Args:
create_anomaly_model_func (func): Method that returns an anomaly model
cell_size (float): Width and height of spatial bin in meter
"""
AnomalyModelBase.__init__(self)
self.CELL_SIZE = cell_size
self.KEY = "%.2f" % self.CELL_SIZE
if fake: self.KEY = "fake_" + self.KEY
self.CREATE_ANOMALY_MODEL_FUNC = create_anomaly_model_func
self.FAKE = fake
m = create_anomaly_model_func()
self.NAME = "SpatialBin/%s/%s" % (m.__class__.__name__.replace(
"AnomalyModel", ""), self.KEY)
def __init__(self,
initial_normal_features=1000,
threshold_learning=300,
threshold_classification=5,
pruning_parameter=0.3):
AnomalyModelBase.__init__(self)
self.NAME += "/%i/%i/%.2f" % (initial_normal_features,
threshold_learning, pruning_parameter)
assert 0 < pruning_parameter < 1, "Pruning parameter out of range (0 < η < 1)"
self.initial_normal_features = initial_normal_features # See reference algorithm variable N
self.threshold_learning = threshold_learning # See reference algorithm variable α
self.threshold_classification = threshold_classification # See reference algorithm variable β
self.pruning_parameter = pruning_parameter # See reference algorithm variable η
self.balanced_distribution = None # Array containing all the "normal" samples
self._mean = None # Mean
self._covI = None # Inverse of covariance matrix
def __generate_model__(self, patches, silent=False):
# Ensure locations are calculated
assert patches.contains_features, "Can only compute patch locations if there are patches"
assert patches.contains_locations, "Can only compute patch locations if there are locations calculated"
# Check if cell size rasterization is already calculated
if not self.KEY in patches.contains_bins.keys(
) or not patches.contains_bins[self.KEY]:
patches.calculate_rasterization(self.CELL_SIZE, self.FAKE)
patches_flat = patches.ravel()
raster = patches.rasterizations[self.KEY]
# Empty grid that will contain the model for each bin
self.models = np.empty(shape=raster.shape, dtype=object)
models_created = 0
with tqdm(desc="Generating models",
total=self.models.size,
file=sys.stderr) as pbar:
for bin in np.ndindex(raster.shape):
indices = raster[bin]
if len(indices) > 0:
model_input = AnomalyModelBase.filter_training(
self, patches_flat[indices])
if model_input.size > 0:
# Create a new model
model = self.CREATE_ANOMALY_MODEL_FUNC(
) # Instantiate a new model
model.__generate_model__(
model_input, silent=silent
) # The model only gets flattened features
self.models[bin] = model # Store the model
models_created += 1
pbar.set_postfix({"Models": models_created})
pbar.update()
return True
def __init__(self):
AnomalyModelBase.__init__(self)
self._var = None # Covariance matrix Ʃ
self._varI = None # Inverse of covariance matrix Ʃ⁻¹
self._mean = None # Mean μ
def __init__(self):
AnomalyModelBase.__init__(self)
self._var = None # Variance σ²
self._mean = None # Mean μ