def createRtree(data):
"""
Creates an R-Tree from the given data
"""
tree = Rtree()
for index, pair in enumerate(data):
tree.insert(index, (pair[3], pair[4]), obj=pair)
return tree
def generate_rtree_from_entities(self):
"""Create an rtree with all entities with bounding rectangles."""
self.bounding_rects = {
id_: e.bounding_rect
for id_, e in self.entities.items() if hasattr(e, 'bounding_rect')
}
self.rtree = Rtree(
(id_, rect, None) for id_, rect in self.bounding_rects.items())
def reset(self, name: Optional[str] = None):
"""Reset index and set name."""
self.name = name
self.id_count = count()
self.entities = {}
self.bounding_rects = {}
self.rtree = Rtree()
self.path_map = PathMap()
self.register_updates = False
self.simulation = Simulation()
self.stats = defaultdict(dict)
self._updates = set()
def _merge_points(points: Dict[int, ConflictPoint], rtree: Rtree):
"""Merge conflict points closer than MERGE_RADIUS."""
curves = set()
merged = set()
for id_, point in points.items():
if id_ in merged:
continue
for other_id in rtree.intersection(
point.point.enclosing_rect(MERGE_RADIUS)):
if other_id == id_:
continue
other = points[other_id]
for curve in other.curves:
curves.add(curve)
curve.replace_conflict_point(other, point)
merged.add(other_id)
rtree.delete(other_id, other.point.bounding_rect)
for id_ in merged:
del points[id_]
for curve in curves:
curve.remove_conflict_point_duplicates()
def _is_minimal_trajectory(self, trajectory: Trajectory,
prior_end_poses: index.Rtree) -> bool:
"""
Determine wheter a trajectory is a minimal trajectory.
Uses an RTree for speedup.
Args:
trajectory: Trajectory
The trajectory to check
prior_end_poses: RTree
An RTree holding the current minimal set of trajectories
Returns
-------
bool
True if the trajectory is a minimal trajectory otherwise false
"""
# Iterate over line segments in the trajectory
for x1, y1, x2, y2, yaw in zip(
trajectory.path.xs[:-1],
trajectory.path.ys[:-1],
trajectory.path.xs[1:],
trajectory.path.ys[1:],
trajectory.path.yaws[:-1],
):
p1 = np.array([x1, y1])
p2 = np.array([x2, y2])
# Create a bounding box search region
# around the line segment
left_bb = min(x1, x2) - self.DISTANCE_THRESHOLD
right_bb = max(x1, x2) + self.DISTANCE_THRESHOLD
top_bb = max(y1, y2) + self.DISTANCE_THRESHOLD
bottom_bb = min(y1, y2) - self.DISTANCE_THRESHOLD
# For any previous end points in the search region we
# check the distance to that point and the angle
# difference. If they are within threshold then this
# trajectory can be composed from a previous trajectory
for prior_end_pose in prior_end_poses.intersection(
(left_bb, bottom_bb, right_bb, top_bb), objects='raw'):
if (self._point_to_line_distance(
p1, p2, prior_end_pose[:-1]) < self.DISTANCE_THRESHOLD
and angle_difference(yaw, prior_end_pose[-1]) <
self.ROTATION_THRESHOLD):
return False
return True
def _fill_neighbors(points: Dict[int, ConflictPoint], rtree: Rtree,
skip_in_same_curve: bool = True):
"""Add conflict points closer than NEIGHBOR_RADIUS as neighbors."""
for id_, point in points.items():
for other_id in rtree.intersection(
point.point.enclosing_rect(NEIGHBOR_RADIUS)):
if other_id == id_:
continue
other = points[other_id]
if skip_in_same_curve and (point.curves & other.curves):
continue
distance_squared = point.point.distance_squared(other.point)
if distance_squared <= NEIGHBOR_RADIUS_SQUARED:
point.neighbors.add(other)
def _build_conflict_points(node: Node, curves: Dict[LaneConnection, Curve]):
"""Create conflict points from lane connection curves.
Fill conflict points in given Curve objects. Merges points that are less
than MERGE_RADIUS apart and add points that are within NEIGHBOR_RADIUS as
neighbors.
"""
id_generator = count()
points: Dict[int, ConflictPoint] = {}
diverge = defaultdict(set)
merge = defaultdict(set)
for (lanes1, curve1), (lanes2, curve2) in combinations(curves.items(), 2):
if lanes1[0] == lanes2[0]:
diverge[lanes1[0]].update((curve1, curve2))
if lanes1[1] == lanes2[1]:
merge[lanes1[1]].update((curve1, curve2))
_add_crossing_conflict_points(id_generator, points, curve1, curve2)
_add_diverge_merge_conflict_points(id_generator, points, diverge, merge)
if len(points) > 1:
rtree = Rtree((id_, p.point.bounding_rect, None)
for id_, p in points.items())
_merge_points(points, rtree)
_fill_neighbors(points, rtree)
for point in points.values():
point.create_lock_order()
# This makes the conflict point positions relative to the node.
point.point = point.point - node.position
for curve in curves.values():
curve.remove_redundant_conflict_point()
# Remove from neighbors points that aren't in any curve
points = {p for _, p in chain.from_iterable(c.conflict_points
for c in curves.values())}
for point in points:
for neighbor in list(point.neighbors):
if neighbor not in points:
point.neighbors.remove(neighbor)
class Rtree2D(object):
"""Wrapper of `rtree.Index` for supporting friendly 2d operations.
Also forces the uniqueness of the `id` parameter, which is different from
the rtree module's behavior.
"""
def __init__(self):
self._index = Rtree()
self._locations = {}
@staticmethod
def to_coords(location):
return (location[0], location[1], location[0], location[1])
def keys(self):
return self._locations.keys()
def get(self, id, objects=False):
return self._locations.get(id)
def set(self, id, location, obj=None):
# Clean up previous value first if any
old = self._locations.get(id)
if old is not None:
self._index.delete(id, self.to_coords(old))
self._locations[id] = location
self._index.insert(id, self.to_coords(location), obj=obj)
def remove(self, id):
self._index.delete(id, self.to_coords(self._locations[id]))
del self._locations[id]
def nearest(self, location, count=1, objects=False, max_distance=None):
ids = self._index.nearest(self.to_coords(location), num_results=count,
objects=objects)
if max_distance is not None:
ids = [id_ for id_ in ids
if distance(self._locations[id_], location) <= max_distance]
return ids
def main(query, train, query_num, qgram_size):
logger = logging.getLogger('search_rtree')
logger.setLevel(logging.DEBUG)
fh = logging.FileHandler('./log/%s' % query)
fh.setLevel(logging.DEBUG)
# create console handler with a higher log level
ch = logging.StreamHandler()
ch.setLevel(logging.DEBUG)
# create formatter and add it to the handlers
formatter = logging.Formatter(
'%(asctime)s - %(name)s - %(levelname)s - %(message)s')
fh.setFormatter(formatter)
ch.setFormatter(formatter)
# add the handlers to the logger
logger.addHandler(fh)
logger.addHandler(ch)
logger.info(
'------------------------- Calculate common Q-grams for query trajectories -------------------------'
)
qgram_tag = 'q_%d' % qgram_size
query_path = './data/processed/%s.txt' % query
rtree_path = './data/interim/%s/my_rtree_%s' % (train, qgram_tag)
logger.info('Query trajectory path: %s' % query_path)
logger.info('Rtree path: %s' % rtree_path)
query_data = load_trajectory(query_path, n=query_num)
logger.info('Load %d query trajectories' % query_num)
qry_qgram, qry_id_list = build_qgram(query_data, qgram_size)
qry_id_dict = build_id_to_key(
qry_id_list) # key: query_id, value: query_key
data_index = Rtree(rtree_path)
conf = SparkConf().setAppName("PythonWordCount").setMaster("local")
sc = SparkContext(conf=conf)
all_data = []
for qry_id, qry_qgrams in qry_qgram.items():
qry_key = qry_id_dict[qry_id]
data = []
for qry_qgram in qry_qgrams:
matches = [
hit.object
for hit in data_index.intersection(qry_qgram, objects=True)
]
matches = set(matches)
data.append(list(matches))
flat_data = [item for sublist in data for item in sublist]
# print(flat_data)
dist_data = sc.parallelize(flat_data)
map_data = dist_data.map(lambda x: (x, 1))
reduce_data = map_data.reduceByKey(lambda a, b: a + b).sortBy(
lambda x: x[1], ascending=False).collect()
# print(reduce_data)
all_data.append([qry_key, reduce_data])
if not os.path.exists('./data/interim/%s' % query):
os.mkdir('./data/interim/%s' % query)
if not os.path.exists('./data/interim/%s/%s' % (query, train)):
os.mkdir('./data/interim/%s/%s' % (query, train))
candidate_traj_path = './data/interim/%s/%s/candidate_trajectory_%s.txt' % (
query, train, qgram_tag)
save_pickle(all_data, candidate_traj_path)
logger.info('Output candidate_trajectory: %s' % candidate_traj_path)
query_id_dict_path = './data/interim/%s/%s/query_id_dict_%s.txt' % (
query, train, qgram_tag)
logger.info('Output query_id_dict: %s' % query_id_dict_path)
save_pickle(qry_id_dict, query_id_dict_path)
gc.collect()
""" Deletes a page """
try:
RTreePage.objects(name=self.name, page=page).delete(safe=True)
except:
returnError.contents.value = self.InvalidPageError
hasData = property( lambda self: RTreePage.objects(name=self.name).first() is not None )
if __name__=='__main__':
settings = Property()
settings.writethrough= True
settings.buffering_capacity=1
storage = MongoStorage('test')
storage.clear()
r = Rtree( storage, properties=settings)
r.add(123,(0,0,1,1))
print "test 1 should be true"
item = list(r.nearest((0,0), 1, objects=True))[0]
print item.id
print r.valid()
print "test 2 should be true"
r.delete(123, (0,0,1,1))
print r.valid()
print "test 3 should be true"
r.clearBuffer()
print r.valid()
RTreePage.objects(name=self.name, page=page).delete(safe=True)
except:
returnError.contents.value = self.InvalidPageError
hasData = property(
lambda self: RTreePage.objects(name=self.name).first() is not None)
if __name__ == '__main__':
settings = Property()
settings.writethrough = True
settings.buffering_capacity = 1
storage = MongoStorage('test')
storage.clear()
r = Rtree(storage, properties=settings)
r.add(123, (0, 0, 1, 1))
print "test 1 should be true"
item = list(r.nearest((0, 0), 1, objects=True))[0]
print item.id
print r.valid()
print "test 2 should be true"
r.delete(123, (0, 0, 1, 1))
print r.valid()
print "test 3 should be true"
r.clearBuffer()
print r.valid()
def __init__(self):
self._index = Rtree()
self._locations = {}
class EntityIndex:
"""Index of spatial entities.
When an entity is added to the index, it gets an unique id and is kept in
a way than can be queried by id or by spatial coordinates.
"""
__slots__ = ('name', 'id_count', 'entities', 'bounding_rects', 'rtree',
'path_map', 'register_updates', 'simulation', 'stats',
'_updates')
extension = 'shelf'
storage_fields = 'id_count', 'entities', 'path_map'
name: str
id_count: count
entities: Dict[int, Entity]
bounding_rects: Dict[int, BoundingRect]
rtree: Rtree
path_map: PathMap
register_updates: bool
simulation: Simulation
# TODO: Define type for stats instead of Any.
stats: Dict[Type, Dict[Any, Any]]
_updates: Set[int]
def __init__(self, name: Optional[str] = None):
self.reset(name)
@property
def filename(self) -> str:
"""Name with extension added."""
if self.name.endswith(f'.{EntityIndex.extension}'):
return self.name
return f'{self.name}.{EntityIndex.extension}'
def reset(self, name: Optional[str] = None):
"""Reset index and set name."""
self.name = name
self.id_count = count()
self.entities = {}
self.bounding_rects = {}
self.rtree = Rtree()
self.path_map = PathMap()
self.register_updates = False
self.simulation = Simulation()
self.stats = defaultdict(dict)
self._updates = set()
def add(self, entity: Entity):
"""Add entity to index."""
if entity.id is not None:
raise ValueError('Entity already has an id.')
entity.id = next(self.id_count)
self.entities[entity.id] = entity
log.debug('[%s] Added %s', __name__, Entity.__repr__(entity))
def add_static(self, entity: Entity):
"""Add entity as static.
Entity may or not have already been added with the `add` method. It
will be added in case it was not already.
A static entity is an entity with geometric information
(`bounding_rect`) that will rarely change. A spatial index is used to
allow for quick spatial queries. These entities are added to the
updated queue when something about them changes. This queue can be
consumed by a front end application with `consume_updates` to update
the representation only when needed.
"""
if entity.id is None:
self.add(entity)
if entity.id in self.bounding_rects:
raise ValueError('Entity already added as static.')
self.bounding_rects[entity.id] = entity.bounding_rect
self.rtree.insert(entity.id, entity.bounding_rect)
self.updated(entity)
def delete(self, entity: Entity):
"""Delete entity from index."""
to_remove = {entity}
while to_remove:
entity = to_remove.pop()
assert self.entities[entity.id] is entity
del self.entities[entity.id]
delete_result = entity.on_delete()
to_remove.update(delete_result.cascade)
for updated in delete_result.updated:
self.updated(updated)
if entity.id in self.bounding_rects:
self.rtree.delete(entity.id, self.bounding_rects[entity.id])
del self.bounding_rects[entity.id]
self.updated(entity)
self.rebuild_path_map()
log.debug('[%s] Removed %s', __name__, entity)
def update_bounding_rect(self,
entity: Entity,
new_rect: Optional[BoundingRect] = None):
"""Change the bounding rectangle of an entity.
Update the bounding rect to `entity.bounding_rect` or to `new_rect` if
it's not None.
"""
assert self.entities[entity.id] is entity
if new_rect is None:
new_rect = entity.bounding_rect
old_rect = self.bounding_rects.get(entity.id, None)
if old_rect is None or old_rect == new_rect:
return
self.rtree.delete(entity.id, old_rect)
self.bounding_rects[entity.id] = new_rect
self.rtree.insert(entity.id, new_rect)
def updated(self, entity: Union[Entity, int]):
"""Mark entity as updated."""
if self.register_updates:
try:
self._updates.add(entity.id)
except AttributeError:
self._updates.add(entity)
def clear_updates(self):
"""Clear entity updates."""
self._updates.clear()
def consume_updates(self) -> Iterator[int]:
"""Get generator that pops and returns updates."""
while self._updates:
yield self._updates.pop()
def generate_rtree_from_entities(self):
"""Create an rtree with all entities with bounding rectangles."""
self.bounding_rects = {
id_: e.bounding_rect
for id_, e in self.entities.items() if hasattr(e, 'bounding_rect')
}
self.rtree = Rtree(
(id_, rect, None) for id_, rect in self.bounding_rects.items())
def load(self, name: Optional[str] = None):
"""Load entities from shelf.
Load enities using the this index name. If a name is passed as
argument, will set the index name before loading.
"""
if name is not None:
self.name = name
with shelve.open(self.filename) as data:
for key in EntityIndex.storage_fields:
log.info('Loading %s', key)
value = data.get(key, None)
if value:
setattr(self, key, value)
log.info('Loaded %s', self.name)
self.generate_rtree_from_entities()
if not hasattr(self, 'path_map'):
self.rebuild_path_map()
def save(self):
"""Save entities to shelf."""
with shelve.open(self.filename) as data:
for key in EntityIndex.storage_fields:
log.info('Saving %s', key)
data[key] = getattr(self, key)
def get_all(self,
of_type: Type[Entity] = None,
where: Callable[[Entity], bool] = None) -> Iterator[Entity]:
"""Get all entities with optional filters."""
def type_filter(entity):
return isinstance(entity, of_type)
filters = []
if of_type is not None:
filters.append(type_filter)
if where is not None:
filters.append(where)
yield from filter(lambda e: all(f(e) for f in filters),
self.entities.values())
def get_at(self,
point: Point,
of_type: Type[Entity] = None,
where: Callable[[Entity], bool] = None) -> List[Entity]:
"""Get entities at given coordinates.
Get a list with entities intersecting the given point. If of_type is
not None, will return only entities of the given type. If where is not
None, where must be a function that receives an Entity and returns True
or False, meaning whether the entity will be returned.
"""
def polygon_filter(entity: Entity) -> bool:
return point_in_polygon(point, entity.polygon)
def type_filter(entity: Entity) -> bool:
return isinstance(entity, of_type)
filters = [polygon_filter]
if of_type is not None:
filters.append(type_filter)
if where is not None:
filters.append(where)
return list(
filter(
lambda e: all(f(e) for f in filters),
map(self.entities.get,
self.rtree.intersection(point.bounding_rect))))
def rebuild_path_map(self):
"""Rebuild the path map, invalidating the old map."""
self.path_map = PathMap()
def main(train, qgram_size):
logger = logging.getLogger('build_rtree')
logger.setLevel(logging.DEBUG)
fh = logging.FileHandler('./log/%s' % train)
fh.setLevel(logging.DEBUG)
# create console handler with a higher log level
ch = logging.StreamHandler()
ch.setLevel(logging.DEBUG)
# create formatter and add it to the handlers
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
fh.setFormatter(formatter)
ch.setFormatter(formatter)
# add the handlers to the logger
logger.addHandler(fh)
logger.addHandler(ch)
logger.info('---------------------------- Build R-tree ----------------------------')
qgram_tag = 'q_%d' % qgram_size
train_path = './data/processed/%s.txt' % train
data = load_trajectory(train_path)
logger.info('Load train trajectory: %s' % train_path)
trajectory, id_list = build_qgram(data, qgram_size)
id_to_key_dict = build_id_to_key(id_list)
# order_key_dict = build_order_dict(id_list)
#save orderId-key mapping
#key: trajectory id in string, value: encoded key
rtree_id_dict_path = './data/interim/%s/rtree_id_dict_%s.txt' % (train, qgram_tag)
save_pickle(id_to_key_dict, rtree_id_dict_path)
logger.info('Output rtree_id_dict: %s' % rtree_id_dict_path)
#key: key, value: trajectory id in string
# filename = '../data/processed/order_key_dict.txt'
# outfile = open(filename,'wb')
# pickle.dump(order_key_dict,outfile)
# outfile.close()
# R-tree constructor
# parameter: 'data_full' is the filename of R-tree storage
# 2 files are created: data_full.dat, data_full.idx
# return: r-tree index
rtree_path = './data/interim/%s/my_rtree_%s' % (train, qgram_tag)
data_idx = Rtree(rtree_path)
logger.info('Output R-tree: %s' % rtree_path)
# put all trajectories into r-tree in the form of bounding box
node_id = 0
start_time = time.time()
for key, qgrams in trajectory.items():
for qgram in qgrams:
# parameters:
# 1. node id
# 2. bounding box(point): (x,y,x,y)
# 3. data inside each node: trajectory's key from order_dict
x = np.around(qgram[0], decimals=5)
y = np.around(qgram[1], decimals=5)
data_idx.insert(node_id, (x, y, x, y), obj=(id_to_key_dict[key]))
node_id += 1
del data_idx
end_time = time.time()
logger.info("exec time: "+str(end_time-start_time))
logger.info('Finished building R-tree')
def searching(feature_vectors_database, feature_vectors_retrieval,
similarity_metric, image_paths, retrieval_number, file,
list_of_parameters, feature_extraction_method, path_database):
'''
feature_vectors: atriutos calculados
labels: label de cada classe
similarity_metric: qual medida utilizar
recuperados as k imagens com menor distancia. Se k = 0, entao o valor eh
setado como sendo o tamanho da classe da imagem
'''
#name to save the pickle file
parameters_name = ""
for parameter in list_of_parameters:
parameters_name = parameters_name + "_" + parameter
file = path_database + "features/sortingRTree" + "_" + feature_extraction_method + parameters_name + '_' + similarity_metric
feature_vectors_retrieval = preprocessing.scale(feature_vectors_retrieval)
if not (os.path.isfile(file + '.dat')):
#normalize signatures
feature_vectors_database = preprocessing.scale(
feature_vectors_database)
# Create a N-Dimensional index
p = index.Property()
p.dimension = feature_vectors_database.shape[1]
idx = index.Index(file, properties=p)
# Create the tree
for i, vector in enumerate(feature_vectors_database):
idx.add(i, vector.tolist())
#save_format = idx.dumps(idx)
#with open(file, 'wb') as handle:
#pickle.dump(save_format, handle)
else:
# Create a N-Dimensional index
p = index.Property()
p.dimension = feature_vectors_database.shape[1]
idx = Rtree(file, properties=p)
#with open(file, 'rb') as handle:
# idx = pickle.load(handle)
# Find closests pair for the first N points
########### debug this part ###########
small_distances = []
for id1, query in enumerate(feature_vectors_retrieval):
nearest = list(idx.nearest(query.tolist(), retrieval_number))
small_distances.append(nearest)
result = []
for cont1, i in enumerate(small_distances):
aux = []
for j in i:
aux.append(image_paths[j])
result.append(aux)
return result
