Venue Distributors

Topic: Distributors
Paths: configs/2021/distributors/school_distributor.yaml, university_distributor.yaml, company_distributor.yaml, hospital_distributor.yaml

Overview

A venue distributor allocates agents to venues and records the assignment on the agent's activity_map. Each YAML file targets one venue_type (school, university, company, hospital). All share the same top-level schema; only the values differ.

Allocation proceeds in phases:

Special cases — mandatory overrides handled before anything else (e.g. boarding-school students matched to their named school).
Priority groups — defined sub-populations allocated in order, optionally with per-area probabilities. Each group may permit overflow.
Normal allocation — remaining eligible agents allocated by the configured strategy and distance rules.
Fallback — agents still unallocated after all phases are handled according to fallback.strategy.

Keys

Key	Description
`venue_type`	Must match a key in `venues_config.yaml`
`activity_map_key`	Key written to `person.activity_map` on assignment
`subset_key`	Subset the agent is added to within the venue
`special_cases`	Priority overrides applied before eligibility filters
`eligibility`	Who is eligible and how they are prioritised
`venue_selection`	How candidate venues are found for each agent
`allocation`	How a venue is chosen from candidates and capacity managed
`settings`	Logging and performance
`fallback`	Behaviour when no eligible venue is found
`validation`	Required attributes checked before allocation
`exports`	Optional CSV reports written after allocation

`venue_type`, `activity_map_key`, `subset_key`

venue_type: "school"
activity_map_key: "primary_activity"
subset_key: "student"

venue_type determines which loaded venues are candidates. activity_map_key is the key under which the assigned venue is stored on person.activity_map; the hospital distributor uses "medical_facility" rather than "primary_activity". subset_key is the subset name added to the venue; omit to skip subset assignment.

`special_cases`

special_cases:
  - name: "boarding_school_students"
    condition:
      person_residence_type: "boarding_school"
    allocation_rule:
      match_by:
        - field: "name"
          source: "person.residence.name"
          target: "venue.name"
          match_type: "exact"
      if_no_match: "error"   # "error" | "warn" | "skip"

Special cases are checked before any eligibility filter and are tried in list order. A matched agent bypasses the normal pipeline. if_no_match controls what happens when no venue matches: "error" halts, "warn" logs and leaves the person unassigned, and "skip" leaves the person unassigned silently. There is no automatic retry through normal allocation.

`eligibility`

eligibility:
  require_unassigned: true
  global_filters:
    - attribute: "age"
      type: "numerical"
      min: 5
      max: 19
    - attribute: "residence.type"
      type: "categorical"
      values: ["household", "boarding_school"]
  exclude:
    households:
      original_pattern: "0 >=0 0 0"
  attributes:
    - name: "sex"
      type: "categorical"
      venue_column: "Gender"
      matching_rules:
        "Mixed": ["male", "female"]
        "Boys": ["male"]
        "Girls": ["female"]
      case_sensitive: false
  priority_allocation:
    enabled: true
    priority_order: "age_desc"
    groups:
      - name: "high_priority_school_age"
        priority: 1
        allow_overflow: true
        search_limits: [20, 70]
        filters:
          - attribute: "age"
            type: "numerical"
            min: 5
            max: 17
        probability_config:
          type: "file"
          file_path:
            - "data/activities/university/EW_university_probabilities.csv"
            - "data/activities/university/SCT_university_probabilities.csv"
            - "data/activities/university/NI_university_probabilities.csv"
          lookup_column: "geo_unit"
          lookup_attribute: "geographical_unit.name"
          probability_column: "prob_uni_18_22"

require_unassigned — when true, skips anyone who already has activity_map_key assigned.

global_filters apply to all phases. Each filter has an attribute path, a type ("numerical" or "categorical"), and type-specific bounds (min/max) or allowed values. Filters are checked in order; list more restrictive filters first for efficiency. Attribute paths may traverse nested objects using dot notation (e.g. residence.type, properties.workplace_sgu).

exclude.households.original_pattern removes agents from households whose original_pattern property matches the given string.

attributes matches agent properties against venue CSV columns. Each entry names a venue_column and a matching_rules dict mapping CSV values to lists of valid agent values. Missing categorical venue values use the engine's fixed Mixed fallback; missing numerical bound columns impose no constraint.

priority_allocation.groups are processed in priority order before normal allocation. allow_overflow: true permits the group to exceed venue capacity whilst still respecting attributes constraints. search_limits is a list of candidate counts tried in sequence (e.g. [20, 70] — try 20 closest, then 70). probability_config optionally samples agents stochastically: type: "file" loads a CSV, matching rows by lookup_column against the agent attribute named by lookup_attribute, and reads allocation probability from probability_column; default is used when the agent's geo unit is not found in the file. priority_order: "age_desc" processes older agents first within each group.

`venue_selection`

venue_selection:
  consider_by: "count"       # "count" | "distance" | "geo_unit"
  count: 10
  search_limits: [20, 50]
  max_distance: 10
  max_distance_unit: "km"    # "km" | "miles" | "meters"
  venue_geo_level: "SGU"     # "SGU" | "MGU" | "LGU"
  person_location_source: "geographical_unit.coordinates"
  respect_capacity: true

consider_by controls how candidate venues are identified. "count" selects the count closest venues. "distance" selects all venues within max_distance. "geo_unit" restricts to venues sharing the agent's geo unit (used by the company distributor, which matches by workplace_sgu rather than residence).

venue_geo_level declares the geography level at which venue coordinates are stored; the engine traverses the hierarchy when agent and venue levels differ.

person_location_source is the attribute path used to read the agent's location. The company distributor sets this to "properties.workplace_sgu" to match agents to companies near their work location rather than their home.

Distances are computed using haversine distance from latitude/longitude pairs.

search_limits gives a fallback candidate sequence for the global pipeline (individual priority groups may specify their own).

`allocation`

allocation:
  strategy: "random"        # "random" | "closest" | "proportional"
  capacity_column: "SchoolCapacity"
  capacity_handling:
    if_missing: "ignore"    # "ignore" | "skip" | "default"
    default_capacity: 1000
    if_zero: "ignore"       # "ignore" | "skip"
  track_capacity: true
  when_full: "exclude"      # "exclude" | "overflow"
  enforce_no_empty_venues: false
  batch_by: "geo_unit"      # "geo_unit" | "none"

strategy selects from the candidate set: "random" picks uniformly; "closest" always picks the nearest; "proportional" weights by inverse distance.

capacity_column names the CSV column holding venue capacity. capacity_handling controls what happens when the column is absent or zero. track_capacity enables runtime tracking so full venues are excluded from subsequent candidates. when_full: "exclude" removes a full venue from the candidate set; "overflow" allows over-capacity assignment (always applies to priority groups with allow_overflow: true).

enforce_no_empty_venues — when true, post-allocation step that moves one agent from the most-populated venue to each empty venue, minimising the number of venues with zero occupancy.

batch_by: "geo_unit" groups agents sharing a geo unit and performs a single spatial query for the batch, reducing the number of distance calculations significantly.

`settings`

settings:
  verbose: true
  use_spatial_index: true

priority controls execution order across all distributors; higher runs first. use_spatial_index builds a KD-tree over venue coordinates for fast nearest-neighbour queries — disable only for very small venue sets or debugging.

`fallback`

fallback:
  strategy: "skip"    # "skip" | "relax_distance" | "relax_capacity" | "assign_closest"
  relax_params:
    max_iterations: 3
    distance_multiplier: 2.0

Applied to any agent still unassigned after all phases. "skip" leaves the agent unassigned. "relax_distance" retries with an expanded search radius, doubling it each iteration up to max_iterations. "relax_capacity" retries ignoring capacity limits. "assign_closest" assigns the nearest venue regardless of all constraints.

The hospital distributor uses "assign_closest" — every agent must be assigned a medical facility.

`validation`

validation:
  required_person_attributes:
    - "age"
    - "geographical_unit"

required_person_attributes — agents missing any of these are skipped before allocation.

Key differences by venue type

Config	`consider_by`	`allocation.strategy`	`fallback.strategy`	Notes
`school_distributor`	`count`	`random`	`skip`	Age + gender constraints; boarding-school special case; priority groups by age band
`university_distributor`	`count`	`proportional`	`skip`	Stochastic priority groups with per-SGU probability file
`company_distributor`	`geo_unit`	`random`	`skip`	Matches by `workplace_sgu`; `work_sector` must match `industry_code`
`hospital_distributor`	`count` (1)	`closest`	`assign_closest`	No eligibility filters; `respect_capacity: false`; everyone assigned

Venue Distributors

Overview

Keys

venue_type, activity_map_key, subset_key

special_cases

eligibility

venue_selection

allocation

settings

fallback

validation