Skip to content

distance

distance #

Distance functions measuring statistical similarity between sessions.

Distance functions measure how "different" two sessions are along a specific dimension. The Void & Cluster algorithm uses these to select captures that maximize diversity across all dimensions.

All distance functions operate on their specific data type (e.g., SolarPosition, Weather, Location). The combine() function handles attribute extraction via keyword arguments.

Available Distance Functions

Type-specific: - solar(sigma_deg): Angular separation on the celestial sphere. Operates on SolarPosition objects. - location(sigma_m): Geographic separation in meters. Operates on cg.Location objects. - weather(...): Multi-dimensional weather difference. Operates on cg.Weather objects. - viewpoint(sigma_deg): Angular separation of camera optical axes. Operates on cg.Pose objects.

Primitive: - scalar(sigma): Simple absolute difference. - angle_degrees(sigma_deg): Circular angle distance in degrees. - angle_radians(sigma_rad): Circular angle distance in radians. - time_of_day(sigma): Circular 24-hour time distance.

Combination: - combine(**kwargs): Euclidean combination of distance functions. The key specifies the attribute path to extract from each session.

Example 
import capturegraph.scheduling as cgsh
from datetime import timedelta

# Single type-specific distance (for use with combine)
solar_dist = cgsh.distance.solar(sigma_deg=2.0)

# Combine multiple distance functions via attribute paths
dist_fn = cgsh.distance.combine(
    solar_angle=cgsh.distance.solar(sigma_deg=2.0),
    location=cgsh.distance.location(sigma_m=100.0),
)

# Mix type-specific and primitive functions
dist_fn = cgsh.distance.combine(
    solar_angle=cgsh.distance.solar(sigma_deg=2.0),
    weather=cgsh.distance.weather(sigma_cloud_cover_ratio=0.3),
    date=cgsh.distance.time_of_day(sigma=timedelta(hours=2)),
)