Source code for aequilibrae.transit.transit

import os
import shutil
import warnings

from aequilibrae.log import logger
from typing import Dict, List

from aequilibrae.project.project_creation import initialize_tables
from aequilibrae.reference_files import spatialite_database
from aequilibrae.transit.lib_gtfs import GTFSRouteSystemBuilder
from aequilibrae.transit.transit_graph_builder import TransitGraphBuilder
from aequilibrae.paths.graph import TransitGraph
from aequilibrae.project.database_connection import database_connection
from aequilibrae.utils.db_utils import read_and_close
from aequilibrae.utils.aeq_signal import SIGNAL
from aequilibrae.utils.interface.worker_thread import WorkerThread
import sqlite3
import pandas as pd


[docs] class Transit(WorkerThread): transit = SIGNAL(object) default_capacities = { 0: [150, 300], # Tram, Streetcar, Light rail 1: [280, 560], # Subway/metro 2: [700, 700], # Rail 3: [30, 60], # Bus 4: [400, 800], # Ferry 5: [20, 40], # Cable tram 11: [30, 60], # Trolleybus 12: [50, 100], # Monorail "other": [30, 60], } default_pces = {0: 5.0, 1: 5.0, 3: 4.0, 5: 4.0, 11: 3.0, "other": 2.0} graphs: Dict[str, TransitGraph] = {} pt_con: sqlite3.Connection
[docs] def __init__(self, project): """ :Arguments: **project** (:obj:`Project`, *Optional*): The Project to connect to. By default, uses the currently active project """ WorkerThread.__init__(self, None) self.project_base_path = project.project_base_path self.logger = logger self.__transit_file = os.path.join(project.project_base_path, "public_transport.sqlite") self.periods = project.network.periods self.create_transit_database() self.pt_con = database_connection("transit")
[docs] def new_gtfs_builder(self, agency, file_path, day="", description="") -> GTFSRouteSystemBuilder: """Returns a ``GTFSRouteSystemBuilder`` object compatible with the project :Arguments: **agency** (:obj:`str`): Name for the agency this feed refers to (e.g. 'CTA') **file_path** (:obj:`str`): Full path to the GTFS feed (e.g. 'D:/project/my_gtfs_feed.zip') **day** (:obj:`str`, *Optional*): Service data contained in this field to be imported (e.g. '2019-10-04') **description** (:obj:`str`, *Optional*): Description for this feed (e.g. 'CTA2019 fixed by John Doe') :Returns: **gtfs_feed** (:obj:`StaticGTFS`): A GTFS feed that can be added to this network """ gtfs = GTFSRouteSystemBuilder( network=self.project_base_path, agency_identifier=agency, file_path=file_path, day=day, description=description, capacities=self.default_capacities, pces=self.default_pces, ) gtfs.signal = self.transit gtfs.gtfs_data.signal = self.transit return gtfs
[docs] def create_transit_database(self): """Creates the public transport database""" if not os.path.exists(self.__transit_file): shutil.copyfile(spatialite_database, self.__transit_file) initialize_tables(self, "transit")
[docs] def create_graph(self, **kwargs) -> TransitGraphBuilder: period_id = kwargs.get("period_id", self.periods.default_period.period_id) graph = TransitGraphBuilder(self.pt_con, period_id, **kwargs) graph.create_graph() self.graphs[period_id] = graph return graph
[docs] def save_graphs(self, period_ids: List[int] = None): # TODO: Support multiple graph saving warnings.warn( "Currently only a single transit graph can be saved and reloaded. Multiple graph support is plan for a future release." ) if period_ids is None: period_ids = [self.periods.default_period.period_id] if len(period_ids) > 1: raise ValueError("Multiple graphs can currently be saved.") for period_id in period_ids: self.graphs[period_id].save()
[docs] def load(self, period_ids: List[int] = None): # TODO: Support multiple graph loading warnings.warn( "Currently only a single transit graph can be saved and reloaded. Multiple graph support is plan for a future release. `period_ids` argument is currently ignored." ) if period_ids is None: period_ids = [self.periods.default_period.period_id] if len(period_ids) > 1: raise ValueError("Multiple graphs can currently be loaded.") self.graphs[period_ids[0]] = TransitGraphBuilder.from_db(self.pt_con, period_ids[0])
[docs] def build_pt_preload(self, start: int, end: int, inclusion_cond: str = "start") -> pd.DataFrame: """Builds a preload vector for the transit network over the specified time period :Arguments: **start** (:obj:`int`): The start of the period for which to check pt schedules (seconds from midnight) **end** (:obj:`int`): The end of the period for which to check pt schedules, (seconds from midnight) **inclusion_cond** (:obj:`str`): Specifies condition with which to include/exclude pt trips from the preload. :Returns: **preloads** (:obj:`pd.DataFrame`): A DataFrame of preload from transit vehicles that can be directly used in an assignment .. code-block:: python >>> project = create_example(project_path, "coquimbo") >>> project.network.build_graphs() >>> start = int(6.5 * 60 * 60) # 6:30 am >>> end = int(8.5 * 60 * 60) # 8:30 am >>> transit = Transit(project) >>> preload = transit.build_pt_preload(start, end) """ return pd.read_sql(self.__build_pt_preload_sql(start, end, inclusion_cond), self.pt_con)
def __build_pt_preload_sql(self, start, end, inclusion_cond): probe_point_lookup = { "start": "MIN(departure)", "end": "MAX(arrival)", "midpoint": "(MIN(departure) + MAX(arrival)) / 2", } def select_trip_ids(): in_period = f"BETWEEN {start} AND {end}" if inclusion_cond == "any": return f"SELECT DISTINCT trip_id FROM trips_schedule WHERE arrival {in_period} OR departure {in_period}" return f""" SELECT trip_id FROM trips_schedule GROUP BY trip_id HAVING {probe_point_lookup[inclusion_cond]} {in_period} """ # Convert trip_id's to link/dir's via pattern_id's return f""" SELECT pm.link as link_id, pm.dir as direction, SUM(r.pce) as preload FROM (SELECT pattern_id FROM trips WHERE trip_id IN ({select_trip_ids()})) as p INNER JOIN pattern_mapping pm ON p.pattern_id = pm.pattern_id INNER JOIN routes r ON p.pattern_id = r.pattern_id GROUP BY pm.link, pm.dir """