Notebook: Quick accessibility analysis using H3pandas#

Chingiz Zhanarbaev, 03.2024

This notebook demonstrates use-case of NxTransit. In this notebook, we will use the library to calculate mean travel time for different parts of Chelyabinsk, Russia using H3 grid.

import pandas as pd
import osmnx as ox
import geopandas as gpd
import h3pandas

import nxtransit as nt

Initialize data#

The first step is to initialize a graph object. Transit graph is a nx.DiGraph object which contains special attributes for handling dynamic nature of GTFS data.

# Setting up the parameters
GTFSpath = "Feeds\Chelyabinsk" # Path to the folder with GTFS files
departure_time_input = "12:00:00" # # Departure time in HH:MM:SS format
day = 'monday' # # Day of the week

transit_graph = nt.feed_to_graph(
    GTFSpath=GTFSpath,
    day_of_week=day,
    departure_time_input=departure_time_input,
    duration_seconds=3600 * 14,
    multiprocessing=True,
)

GTFS feed is valid.
151276 of 242644 trips retained
Building graph in parallel

INFO - Transit graph created
INFO - Loading OSM graph via OSMNX
INFO - Street network graph created
INFO - Nodes: 87834, Edges: 239108

Prepare buildings geodataframe using OSMnx#

tags = {"building": True}
buildings = ox.features_from_place("Chelyabinsk", tags)
building_centroids = gpd.GeoDataFrame(buildings, geometry=buildings.geometry.centroid)

Aggregate OSM buildings with H3 cells using `h3pandas`#

Resulting geodataframe will contain only cells that have buildings in them.

building_centroids_h3 = building_centroids.h3.geo_to_h3_aggregate(9, return_geometry=True)
# Filter unnecessary columns
building_centroids_h3 = building_centroids_h3[["geometry"]]

building_centroids_h3.explore(tooltip="h3_09", tiles = "cartodbpositron")

Make this Notebook Trusted to load map: File -> Trust Notebook

Calculate OD-cost matrix#

Snap H3 centroids to network#

Add id field to the polygons, which will be used as a node id in the graph

building_centroids_h3['id'] = building_centroids_h3.index
hex_centroids = nt.create_centroids_dataframe(building_centroids_h3)
nt.snap_points_to_network(transit_graph, hex_centroids)

Perform OD-cost matrix calculation#

# Create list of nodes for OD calculation
nodes = list(hex_centroids["origin_id"])
od_matrix = nt.calculate_od_matrix_parallel(graph=transit_graph, nodes=nodes, departure_time=3600 * 18, num_processes=6)

Calculating the OD using 6 processes
Time elapsed: 567.5380474000121

Group OD-cost matrix by source node and calculate the average travel time to all other nodes

od_group = od_matrix.groupby('source_node')['travel_time'].mean().reset_index()

Merge data from OD-cost matrix with the H3 geodataframe

df = pd.merge(building_centroids_h3, od_group, left_on="id", right_on="source_node")
df = df[["geometry", "travel_time"]]

Plot results using folium#

df.explore(
    column="travel_time",
    cmap="RdYlGn_r",
    tiles = "cartodbpositron",
    legend=True,
    tooltip="travel_time",
    legend_kwds=dict(colorbar=False),
    scheme="Quantiles",
    k=6,
)