Map example

This example builds a small street network, resolves two missing segment travel times, and plots the geographic network beside its travel-time deformation. It is self-contained and does not download map tiles or data.

Run it

Clone the repository and run the checked-in example with the plotting extra:

cd spatialdeform
uv run --extra plot python examples/map_example.py

The script writes docs/assets/map-example.svg. Its essential workflow is:

import numpy as np
from spatialdeform import SpatialDeformer

model = SpatialDeformer(
    cost="travel_time_s",
    missing_cost="median",
    impute_by="road_class",
    geo_weight=0.08,
)
deformed = model.fit_transform(edges)

status = np.where(model.edge_mask_, "observed", "dropped")
status[np.isin(edges.index, model.imputed_edges_)] = "imputed"

The complete data construction and plotting code is in examples/map_example.py.

Read the result

The orange segments had no observed travel time. They remain visible and are used in fitting only after their values are resolved. Here the local-road segment receives the local-road median; the missing service-road group has no observations, so it receives the global median.

for edge_index in model.imputed_edges_:
    print(edge_index, model.costs_[edges.index.get_loc(edge_index)])

Keep that distinction in analytical exports too. The fitted audit attributes are aligned with the input rows:

Attribute	Meaning
costs_	resolved costs aligned with input; dropped positions stay missing
edge_mask_	whether the row participated in graph fitting
imputed_edges_	input index labels whose values were filled
dropped_edges_	input index labels omitted from graph fitting
n_edges_used_	number of graph-fitting edges

Try another policy

Use missing_cost="length" with a speed in projected distance units per cost unit, or missing_cost="drop" if the remaining network is still connected. See Input data for policy details and failure modes.