# Logan Noel (github.com/lmnoel)
#
# ©2017-2019, Center for Spatial Data Science
import time
import logging
import os
import scipy.spatial
from geopy import distance
import pandas as pd
from numpy import issubdtype, integer, signedinteger
from spatial_access.MatrixInterface import MatrixInterface
from spatial_access.NetworkInterface import NetworkInterface
from spatial_access.Configs import Configs
from spatial_access.SpatialAccessExceptions import PrimaryDataNotFoundException
from spatial_access.SpatialAccessExceptions import SecondaryDataNotFoundException
from spatial_access.SpatialAccessExceptions import UnableToParsePrimaryDataException
from spatial_access.SpatialAccessExceptions import UnableToParseSecondaryDataException
from spatial_access.SpatialAccessExceptions import UnknownModeException
from spatial_access.SpatialAccessExceptions import InsufficientDataException
from spatial_access.SpatialAccessExceptions import DuplicateInputException
from spatial_access.SpatialAccessExceptions import WriteTMXFailedException
from spatial_access.SpatialAccessExceptions import WriteCSVFailedException
from spatial_access.SpatialAccessExceptions import ImproperIndecesTypeException
class TransitMatrix:
"""
Compute transit matrices at scale.
"""
def __init__(
self,
network_type,
primary_input=None,
secondary_input=None,
read_from_file=None,
primary_hints=None,
secondary_hints=None,
debug=False,
configs=None):
"""
Args:
network_type: string, one of {'walk', 'bike', 'drive', 'otp'}.
primary_input: string, csv filename.
secondary_input: string, csv filename (omit to calculate an NxN matrix on
the primary_input).
read_from_file: string, tmx or csv filename.
primary_hints: dictionary, map column names to expected values.
secondary_hints: dictionary, map column names to expected values.
debug: boolean, enable to see more detailed logging output.
configs: defaults to None, else pass in an instance of Configs to override
default values.
Raises:
UnknownModeException: If the network type is unknown.
DuplicateInputException: If the same file is given as primary_input
and secondary_input. To compute symmetric matrices (NxN), leave the
secondary input field blank.
InsufficientDataException: If neither a source data file (csv) nor a
transit matrix file (tmx) is supplied.
"""
# arguments
self.network_type = network_type
self.primary_input = primary_input
self.secondary_input = secondary_input
self.primary_hints = primary_hints
self.secondary_hints = secondary_hints
# member variables
self.primary_data = None
self.secondary_data = None
# start the logger
self.logger = None
self.set_logging(debug)
# instantiate interfaces
if isinstance(configs, Configs):
self.logger.debug("set custom config")
self.configs = configs
else:
self.configs = Configs()
self._network_interface = NetworkInterface(network_type, logger=self.logger,
disable_area_threshold=self.configs.disable_area_threshold)
self.matrix_interface = MatrixInterface(logger=self.logger,
require_extended_range=self.configs.require_extended_range)
if network_type not in {'drive', 'walk', 'bike', 'otp'}:
raise UnknownModeException(network_type)
if self.primary_input == self.secondary_input and self.primary_input is not None:
raise DuplicateInputException("Gave duplicate inputs: {}".format(self.primary_input))
# need to supply either:
if primary_input is None and read_from_file is None:
raise InsufficientDataException()
if read_from_file:
self.matrix_interface.read_file(read_from_file)
if network_type == 'otp':
self.matrix_interface.read_otp(primary_input)
def set_logging(self, debug):
"""
Set the logging level.
Args:
debug: enable for increased details
in logs.
"""
if debug:
logging.basicConfig(level=logging.DEBUG)
else:
logging.basicConfig(level=logging.INFO)
self.logger = logging.getLogger(__name__)
self.logger.debug("Running in debug mode")
@staticmethod
def _get_output_filename(keyword, extension):
"""
Args:
keyword: the file's keyword.
extension: the files's type.
Returns: unique filename.
"""
if not os.path.exists("data/matrices/"):
os.makedirs("data/matrices/")
if extension is None:
filename = 'data/matrices/{}_0'.format(keyword)
else:
filename = 'data/matrices/{}_0.{}'.format(keyword, extension)
counter = 1
while os.path.isfile(filename):
if extension is None:
filename = 'data/matrices/{}_{}'.format(
keyword, counter)
else:
filename = 'data/matrices/{}_{}.{}'.format(
keyword, counter, extension)
counter += 1
return filename
@staticmethod
def _get_type_of_series(series):
"""
Returns: type of the series (int16, int32, int64, int128 or str)
Raises:
ImproperIndecesTypeException: If the series
is not one of the expected types.
"""
if type(series[0]) == str:
return str
elif issubdtype(series.dtype, integer) or issubdtype(series.dtype, signedinteger):
return integer
raise ImproperIndecesTypeException(str(series.dtype))
def _parse_csv(self, primary):
"""
Load source data from .csv. Identify lon, lon and id columns.
Args:
primary: boolean, true if loading primary data.
Raises:
UnableToParsePrimaryDataException: The user's supplied
mapping to column names failed.
UnableToParseSecondaryDataException: The user's supplied
mapping to column names failed.
"""
if primary:
filename = self.primary_input
else:
filename = self.secondary_input
source_data = pd.read_csv(filename)
source_data_columns = source_data.columns.values
# extract the column names
lon = ''
lat = ''
idx = ''
skip_user_input = False
# use the column names if we already have them
if primary and self.primary_hints:
lon = self.primary_hints['lon']
lat = self.primary_hints['lat']
idx = self.primary_hints['idx']
skip_user_input = True
elif not primary and self.secondary_hints:
lon = self.secondary_hints['lon']
lat = self.secondary_hints['lat']
idx = self.secondary_hints['idx']
skip_user_input = True
if not skip_user_input:
print('The variables in your data set are:')
for var in source_data_columns:
print('> ', var)
while lon not in source_data_columns:
lon = input('Enter the longitude coordinate: ')
while lat not in source_data_columns:
lat = input('Enter the latitude coordinate: ')
while idx not in source_data_columns:
idx = input('Enter the index name: ')
# drop nan lines
pre_drop = len(source_data)
source_data.dropna(subset=[lon, lat], axis='index', inplace=True)
dropped_lines = pre_drop - len(source_data)
keyword = "rows" if primary else "columns"
self.logger.debug(
'Total number of {} in the dataset: {}'.format(keyword, pre_drop))
if dropped_lines > 0:
self.logger.warning(
"Rows dropped due to missing latitude or longitude values: %d", dropped_lines)
# set index and clean
if primary:
self.matrix_interface.primary_ids_are_string = self._get_type_of_series(source_data[idx]) == str
else:
self.matrix_interface.secondary_ids_are_string = self._get_type_of_series(source_data[idx]) == str
source_data.set_index(idx, inplace=True)
source_data.rename(columns={lon: 'lon', lat: 'lat'}, inplace=True)
if primary:
self.primary_data = source_data[['lon', 'lat']]
self.primary_hints = {'idx': idx, 'lon': lon, 'lat': lat}
else:
self.secondary_data = source_data[['lon', 'lat']]
self.secondary_hints = {'idx': idx, 'lon': lon, 'lat': lat}
def _load_inputs(self):
"""
Load one input file if the user wants a symmetric
transit matrix, or two for an asymmetric matrix.
Raises:
PrimaryDataNotFoundException: Primary data isn't found.
SecondaryDataNotFoundException: Secondary data isn't found.
"""
if not os.path.isfile(self.primary_input):
self.logger.error("Unable to find primary csv.")
raise PrimaryDataNotFoundException("Unable to find primary csv")
if self.secondary_input:
if not os.path.isfile(self.secondary_input):
self.logger.error("Unable to find secondary csv.")
raise SecondaryDataNotFoundException("Unable to find secondary csv")
else:
self.matrix_interface.secondary_ids_are_string = self.matrix_interface.primary_ids_are_string
try:
self._parse_csv(True)
except KeyError:
raise UnableToParsePrimaryDataException()
if self.secondary_input:
try:
self._parse_csv(False)
except KeyError:
raise UnableToParseSecondaryDataException()
def _reduce_node_indeces(self):
"""
Map the network indeces to location.
Returns:
dictionary of {node index : node location}
"""
simple_node_indeces = {}
for position, id_ in enumerate(self._network_interface.nodes['id']):
simple_node_indeces[id_] = position
return simple_node_indeces
def _parse_network(self):
"""
Cleans and generates the city network.
"""
start_time = time.time()
edges = self._network_interface.edges
if self.configs.use_meters:
edges['edge_weight'] = edges['distance']
elif self.network_type == 'walk':
edges['edge_weight'] = edges['distance'] / self.configs._get_walk_speed() \
+ self.configs.walk_node_penalty
elif self.network_type == 'bike':
edges['edge_weight'] = edges['distance'] / self.configs._get_bike_speed() \
+ self.configs.bike_node_penalty
elif self.network_type == 'drive':
driving_cost_matrix = self.configs._get_driving_cost_matrix()
edges = pd.merge(edges, driving_cost_matrix, how='left', left_on='highway', right_index=True)
edges['unit_cost'].fillna(self.configs._get_default_drive_speed(), inplace=True)
edges['edge_weight'] = edges['distance'] / edges['unit_cost'] + self.configs.drive_node_penalty
if self.network_type == 'walk' or self.network_type == 'bike':
edges['is_bidirectional'] = True
elif self.network_type == 'drive':
edges['is_bidirectional'] = edges['oneway'] != "yes"
simple_node_indeces = self._reduce_node_indeces()
edges['from_loc'] = edges['from'].map(simple_node_indeces)
edges['to_loc'] = edges['to'].map(simple_node_indeces)
edges['edge_weight'] = edges['edge_weight'].astype('int16')
from_column = list(edges['from_loc'])
to_column = list(edges['to_loc'])
edge_weight_column = list(edges['edge_weight'])
is_bidirectional_column = list(edges['is_bidirectional'])
self.matrix_interface.add_edges_to_graph(from_column, to_column, edge_weight_column,
is_bidirectional_column)
time_delta = time.time() - start_time
self.logger.debug("Prepared raw network in {:,.2f} seconds".format(time_delta))
def _match_to_nearest_neighbor(self, is_primary=True, is_also_secondary=False):
"""
Map each vertex in the user's data set to a vertex in
the underlying osm network.
Args:
is_primary: true if this is the primary dataset.
is_also_secondary: true if this is also acting as the secondary dataset.
"""
if is_primary:
data = self.primary_data
else:
data = self.secondary_data
nodes = self._network_interface.nodes[['x', 'y']]
start_time = time.time()
# make a kd tree in the lat, long dimension
node_array = nodes.values
kd_tree = scipy.spatial.cKDTree(node_array)
unit_cost = 1
if self.configs.use_meters:
unit_cost = 1
elif self.network_type == 'drive':
unit_cost = self.configs._get_default_drive_speed()
elif self.network_type == 'walk':
unit_cost = self.configs._get_walk_speed()
elif self.network_type == 'bike':
unit_cost = self.configs._get_bike_speed()
else:
assert False, "Unknown type"
# map each node in the source/dest data to the nearest
# corresponding node in the OSM network
# and write to file
for row in data.itertuples():
origin_id, origin_x, origin_y = row
latlong_diff, node_loc = kd_tree.query([origin_x, origin_y], k=1)
node_number = nodes.index[node_loc]
origin_location = (origin_y, origin_x)
closest_node_location = (nodes.loc[node_number].y,
nodes.loc[node_number].x)
# keep track of nodes that are used to snap a user data point
edge_distance = distance.distance(origin_location, closest_node_location).m
edge_weight = int(edge_distance / unit_cost)
if is_primary:
self.matrix_interface.add_user_source_data(network_id=node_loc,
user_id=origin_id,
weight=edge_weight,
is_also_dest=is_also_secondary)
else:
self.matrix_interface.add_user_dest_data(network_id=node_loc,
user_id=origin_id,
weight=edge_weight)
time_delta = time.time() - start_time
self.logger.debug(
'Nearest Neighbor matching completed in {:,.2f} seconds'.format(time_delta))
def write_csv(self, outfile=None):
"""
Write the transit matrix to csv.
Note: Use write_tmx (as opposed to this method) to
save the transit matrix unless exporting for external use.
Arguments:
outfile: optional filename.
Raises:
WriteCSVFailedException: filename does not have correct extension.
"""
if not outfile:
outfile = self._get_output_filename(self.network_type, extension='csv')
if '.csv' not in outfile:
raise WriteCSVFailedException('given filename does not have the correct extension (.csv)')
self.matrix_interface.write_csv(outfile)
def getRowIds(self):
"""
Get row ids from transition matrix
"""
return self.matrix_interface.getRowIds()
def getColIds(self):
"""
Get col ids from transition matrix
"""
return self.matrix_interface.getColIds()
def write_tmx(self, outfile=None):
"""
Write the transit matrix to tmx.
Note: Use this method (as opposed to write_csv) to
save the transit matrix unless exporting data for
external use.
Arguments:
outfile: optional filename.
Raises:
WriteTMXFailedException: filename does not have correct extension.
"""
if not outfile:
outfile = self._get_output_filename(self.network_type, extension='tmx')
if '.tmx' not in outfile:
raise WriteTMXFailedException('given filename does not have the correct extension (.tmx)')
self.matrix_interface.write_tmx(outfile)
def prefetch_network(self):
"""
Fetch and cache the osm network.
"""
self._load_inputs()
self.logger.debug("Fetching network (%s) with epsilon: %f",
self.network_type, self.configs.epsilon)
self._network_interface.load_network(self.primary_data,
self.secondary_data,
self.secondary_input is not None,
self.configs.epsilon)
@staticmethod
def clear_cache():
"""
Clear the network cache.
"""
NetworkInterface.clear_cache()
def _is_compressible(self):
"""
Returns: true if the transit matrix can be compressed by
half without losing any data.
"""
return self._is_symmetric() and self.network_type in {'walk', 'bike'}
def _is_symmetric(self):
"""
Returns: true if the transit matrix is NxN, that is, has
the same origins and destinations.
"""
return self.secondary_input is None
def process(self):
"""
- Load the users's data.
- Fetch the osm network.
- Parse the network.
- Calculate transit matrix.
Raises:
AssertionError: if this method is called on an OTP-matrix.
"""
assert self.network_type != 'otp', 'no need to call process for an otp matrix'
start_time = time.time()
self.prefetch_network()
rows = len(self.primary_data)
if self.secondary_input is None:
cols = rows
self.matrix_interface.secondary_ids_are_string = self.matrix_interface.primary_ids_are_string
else:
cols = len(self.secondary_data)
self.matrix_interface.prepare_matrix(is_symmetric=self._is_symmetric(),
is_compressible=self._is_compressible(),
rows=rows,
columns=cols,
network_vertices=self._network_interface.number_of_nodes())
if self.secondary_input:
self._match_to_nearest_neighbor(is_primary=True, is_also_secondary=False)
self._match_to_nearest_neighbor(is_primary=False, is_also_secondary=False)
else:
self._match_to_nearest_neighbor(is_primary=True, is_also_secondary=True)
self._parse_network()
# offload primary and secondary input data frames because we don't need them anymore
self.primary_input = None
self.secondary_input = None
self.matrix_interface.build_matrix()
time_delta = time.time() - start_time
self.logger.info('All operations completed in {:,.2f} seconds'.format(time_delta))
