geoflowkit 0.3.0

A package for geospatial flow analysis and visualization

GeoFlowKit

A Python package for handling and analyzing geographical flow data, extending pandas and geopandas with flow-specific operations.

Overview

GeoFlowKit provides FlowSeries and FlowDataFrame types, which are subclasses of pandas.Series and pandas.DataFrame respectively. They are designed to work with flow data consisting of origin-destination (OD) pairs, similar to how geopandas.GeoSeries and geopandas.GeoDataFrame work with geometries.

Installation

pip install geoflowkit

Or install from source:

pip install .

Dependencies

• shapely
• numpy
• pandas
• geopandas >= 1.0.1
• matplotlib
• scikit-learn
• tqdm
• numba
• networkx >= 2.6

Quick Start

Creating Flow Objects

import numpy as np
from geoflowkit import Flow, FlowSeries, FlowDataFrame

# Create a single Flow (origin-destination pair)
flow = Flow([[0, 0], [1, 1]])

# Access origin and destination points
print(flow.o)  # POINT (0 0)
print(flow.d)  # POINT (1 1)

Creating FlowSeries

# From a list of Flow objects
fs = FlowSeries([
    Flow([[0, 0], [1, 1]]),
    Flow([[1, 1], [2, 2]]),
    Flow([[2, 2], [3, 3]])
], crs="EPSG:4326")

# From coordinate arrays using flows_from_od
from geoflowkit import flows_from_od

o_points = np.array([[0, 0], [1, 1], [2, 2]])
d_points = np.array([[1, 1], [2, 2], [3, 3]])
fs = flows_from_od(o_points, d_points, crs="EPSG:4326")

Creating FlowDataFrame

# Create a FlowDataFrame with attributes
data = {
    'id': [1, 2, 3],
    'value': [10, 20, 30],
    'geometry': fs
}
fdf = FlowDataFrame(data, crs="EPSG:4326")
print(fdf)

Reading Data from Files

# Read from CSV (specify origin/destination columns)
fdf = read_csv(
    'flow_data.csv',
    use_cols=['ox', 'oy', 'dx', 'dy'],
    crs='EPSG:4326'
)

# Read from GeoPackage
fdf = read_file('flow_data.gpkg', layer='flows')

Core Features

Flow Properties

# Access origin and destination points
origins = fdf.o  # GeoSeries of origin points
destinations = fdf.d  # GeoSeries of destination points

# Flow length and angle
lengths = fdf.length  # Distance from origin to destination
angles = fdf.angle    # Direction of flow (radians)

# Flow density and volume
density = fdf.density  # Flows per unit area
volume = fdf.volume    # Total bounding area

Flow Metrics

# Calculate pairwise distances between flows
from geoflowkit import pairwise_distances

dist_matrix = pairwise_distances(fdf, distance='max')

# Calculate local density of flow
from geoflowkit import k_neighbor_distances, snn_distance

k_dists = k_neighbor_distances(fdf, k=2)
snn_dist = snn_distance(fdf, k=8)

# Calculate disorder of flows
from geoflowkit import flow_entropy, flow_divergence

entropy = flow_entropy(fdf)
div = flow_divergence(fdf, n_directions=6)

Spatial Operations

# Clip flows within a polygon
clipped = fdf.clip(polygon_mask)

# Select flows within bounds
within_bounds = fdf.within(bounds_box)

# Calculate the distance with others
dist_series = fdf.distance(flow)
dist_series = fdf.distance(other_fdf)

Spatial Clustering Scale Detection (K/L Functions)

# Calculate K function for spatial clustering
from geoflowkit import k_func, l_func

r_list, kr_list = k_func(fdf, dr=0.1, k=1)
r_list, lr_list = l_func(fdf, dr=0.1, k=1)

# Local L function for individual flows
from geoflowkit import local_l_func

llrs = local_l_func(fdf, r=0.5)

Grid Aggregation

# Divide study area into grid and aggregate flows
gridded = fdf.to_grid(delta_x=0.1, delta_y=0.1)

Visualization

# Plot flows as arrows
ax = fdf.plot(kind='arrow', column='value')

# Plot FlowSeries
ax = fs.plot()

OD Matrix & MapTrix

from geoflowkit.visualization import ODMatrixVisualizer, MapTrixVisualizer

# OD Matrix heatmap (colour = flow count)
ODMatrixVisualizer(
    origin_zones=border, zone_id_col='Name',
    weight='count', cmap='OrRd',
).fit_plot(fdf, ax=ax)

# OD Matrix with proportional circles (size = flow length)
ODMatrixVisualizer(
    origin_zones=border, zone_id_col='Name',
    weight='count', size_weight='length',
).fit_plot(fdf, ax=ax)

# MapTrix — rotated matrix + origin/destination maps
MapTrixVisualizer(
    origin_zones=border, zone_id_col='Name',
    weight='count', size_weight='length',
    out_title='Outflow', in_title='Inflow',
).fit_plot(fdf, figsize=(16, 9))

Flow Clustering

# K-medoid clustering
from geoflowkit import kmedoid

labels = kmedoid(fdf, n_clusters=5)

# DBSCAN clustering
from geoflowkit import dbscan

labels = dbscan(fdf, eps=0.5, min_samples=5)

# K-Means clustering (virtual flow centers)
from geoflowkit import kmeans

labels = kmeans(fdf, n_clusters=5, distance='max', random_state=42)

# Or use the class API to inspect cluster centers
from geoflowkit import KMeansFlow

km = KMeansFlow(n_clusters=5, distance='max', random_state=42).fit(fdf)
print(km.cluster_centers_)  # FlowSeries of virtual flow centers

Community Detection

Community detection algorithms identify groups of closely connected zones in flow networks. These algorithms first build a flow network graph from the FlowDataFrame, then detect communities using modularity optimization.

# CNM (Clauset-Newman-Moore) algorithm
from geoflowkit import cnm

labels = cnm(fdf, zone_method='grid', cell_size=1000)

# Louvain algorithm
from geoflowkit import louvain

labels = louvain(fdf, zone_method='grid', cell_size=1000, seed=42)

# STOCS (Spatial Tabu Optimization for Community Structure)
from geoflowkit import stocs

labels = stocs(
    fdf, zone_method='grid', cell_size=1000,
    spatial_weight=0.5, tabu_tenure=15
)

Zone methods: 'grid' (regular grid), 'aggregate' (unique OD pairs), 'gdf' (external GeoDataFrame), 'custom' (user function).

# Custom zone function example
import numpy as np

def my_zones(fdf):
    origins = np.array([[p.x, p.y] for p in fdf.o])
    destinations = np.array([[p.x, p.y] for p in fdf.d])
    o_zones = (origins[:, 0] > 0).astype(int) * 2 + (origins[:, 1] > 0).astype(int)
    d_zones = (destinations[:, 0] > 0).astype(int) * 2 + (destinations[:, 1] > 0).astype(int)
    zone_centroids = {i: (origins[o_zones == i].mean(axis=0)) for i in range(4) if (o_zones == i).any()}
    return o_zones.astype(int), d_zones.astype(int), zone_centroids

labels = louvain(fdf, zone_method='custom', zone_func=my_zones)

Output: Flow-level labels where -1 indicates cross-community flows (origin and destination belong to different communities).

Manifold Learning (FTSNE)

from geoflowkit import FTSNE

# Global interpretability (separate O and D)
transformer = FTSNE(perplexity=200, learning_rate='auto')
X_embedded = transformer.fit_transform(
    fdf, identity={'o': 0, 'd': 1}
)

# Local interpretability (union O and D)
X_embedded = transformer.fit_transform(
    fdf, union={('o', 'd'): (0, 1)}
)

Location Centrality (I-index)

The I-index quantifies the irreplaceability of a location based on flows, combining flow volume and flow length into a single metric following the H-index principle.

from geoflowkit.spatial import i_index

# Calculate I-index for each zone
result = i_index(fdf, zones)

# Using origin points instead of destination
result = i_index(fdf, zones, od_type='o')

# With custom alpha parameter
result = i_index(fdf, zones, alpha=1000.0)

I-index definition: The I-index of a location is the maximum value of i such that at least i flows with a length of at least α × i meters have reached this location. Higher values indicate more irreplaceable locations that attract many long-distance flows.

Examples

Jupyter notebook examples are available in the examples/ folder:

• basic_usage.ipynb - Basic usage of Flow, FlowSeries, and FlowDataFrame
• clustering.ipynb - Flow clustering and community detection
• kl_function.ipynb - K/L functions for spatial clustering detection
• visualization.ipynb - OD Matrix, MapTrix, and FTSNE visualizations
• centrality.ipynb - I-index for location irreplaceability

API Reference

Classes

• Flow: Geometry object representing an origin-destination pair
• FlowSeries: pandas Series subclass for storing Flow objects
• FlowDataFrame: pandas DataFrame subclass with Flow geometry column
• KMedoidFlow: K-medoid clustering for flow data
• DBSCANFlow: DBSCAN clustering for flow data
• KMeansFlow: K-Means clustering for flow data (virtual flow centers)
• CNMFlow: Clauset-Newman-Moore community detection
• LouvainFlow: Louvain community detection
• STOCSFlow: Spatial Tabu Optimization for Community Structure
• FTSNE: A Variant of t-SNE for Flow Data
• ODMatrixVisualizer: OD matrix heatmap visualization
• MapTrixVisualizer: MapTrix layout (matrix + maps + guide lines)

Key Functions

• flows_from_od(o, d, crs=None): Create FlowSeries from coordinate arrays
• flows_from_geometry(geometry, crs=None): Create FlowSeries from geometry objects
• read_csv(file_path, use_cols, crs=None, **kwargs): Read flow data from CSV
• read_file(file_path, **kwargs): Read flow data from vector file
• pairwise_distances(fdf, distance='max', ...): Calculate flow distance matrix
• k_neighbor_distances(fdf, k, distance='max', ...): K-order nearest neighbor distances
• snn_distance(fdf, k, ...): Shared nearest neighbor distance
• flow_entropy(fdf, cell_area=None, ...): Flow space entropy
• flow_divergence(fdf, n_directions=6, ...): Flow directional entropy
• k_func(fdf, dr, k=1, distance='max', ...): K function for spatial clustering detection
• l_func(fdf, dr, k=1, distance='max', ...): L function for spatial clustering detection
• local_l_func(fdf, r, distance='max', ...): Local L function for individual flows
• kmedoid(fdf, n_clusters=5, ...): K-medoid clustering for flows
• dbscan(fdf, eps=0.5, min_samples=5, ...): DBSCAN clustering for flows
• kmeans(fdf, n_clusters=8, distance='max', ...): K-Means clustering for flows (virtual flow centers)
• cnm(fdf, zone_method='grid', cell_size=..., ...): CNM community detection
• louvain(fdf, zone_method='grid', cell_size=..., seed=..., ...): Louvain community detection
• stocs(fdf, zone_method='grid', cell_size=..., spatial_weight=0.5, ...): STOCS community detection
• i_index(fdf, zones, alpha=None, od_type='d', ...): I-index for location irreplaceability
• second_order_density(fdf, ...): Second-order density of flows
• FlowSeries / FlowDataFrame methods:
  • plot(): Render flows as arrows using matplotlib quiver
  • within(mask): Select flows whose origin and destination are both inside mask
  • clip(mask): Clip flows to a mask polygon
  • distance(other, distance='max', ...): Calculate distance to another flow or FlowSeries
  • to_grid(delta_x=None, delta_y=None, ...): Divide study area into grid and aggregate flows (FlowDataFrame only)
  • to_crs(crs): Transform CRS of flows
  • set_crs(crs, allow_override=False): Set CRS without transforming geometries

License

GeoFlowKit is licensed under the MIT License.

Contact

For questions or feedback: [e-mail]