Skip to content

bindings.md

Latest commit

 

History

History
171 lines (140 loc) · 4.17 KB

bindings.md

File metadata and controls

171 lines (140 loc) · 4.17 KB
Raw

Snapshot Bindings Registry

Snapshot bindings registry defines which snapshot of each specified dataset and feature set will be used by a script that uses these configs.

It allows the use of specific snapshots, hence enabling reproducibility.

Unlike some of the other configuration types, snapshot bindings are centralized in a single configuration file and selected dynamically at runtime of a given pipeline (likely freeze.py, search.py, or train.py).

1. Location

Snapshot bindings are defined in a single file:

configs/snapshot_bindings_registry/bindings.yaml

This file contains all of the defined snapshots of snapshot bindings.

2. Configuration Structure

The configuration is organized hierarchically

General structure:

snapshot_id
  └─ snapshot_bindings

Example:

2026-03-20T02-54-47_61509023:
  datasets:
    hotel_bookings:
      v1:
        snapshot: 2026-03-17T03-26-25_4e817323
      v2:
        snapshot: 2026-03-17T03-26-27_c04ca84b
  feature_sets:
    booking_context_features:
      v1:
        snapshot: 2026-03-17T06-41-04_530962ff
      v2:
        snapshot: 2026-03-17T04-27-05_b9099899
      v3:
        snapshot: 2026-03-17T04-27-07_6ecc2061
      v4:
        snapshot: 2026-03-17T04-27-08_e45a7462
      v5:
        snapshot: 2026-03-17T04-27-09_72290e7c
      v6:
        snapshot: 2026-03-17T04-27-10_f299b9d2
      v7:
        snapshot: 2026-03-17T04-27-12_eba7203e
    channel_and_agent_features:
      v1:
        snapshot: 2026-03-17T04-27-24_d2b1a0df
      v2:
        snapshot: 2026-03-17T04-27-25_1bd99237
      v3:
        snapshot: 2026-03-17T04-27-27_e73d2e8c
    customer_history_features:
      v1:
        snapshot: 2026-03-17T04-27-13_47665433
      v2:
        snapshot: 2026-03-17T04-27-14_1881a909
      v3:
        snapshot: 2026-03-17T04-27-15_c7be8caf
      v4:
        snapshot: 2026-03-17T04-27-17_1c0339c3
    pricing_party_features:
      v1:
        snapshot: 2026-03-17T04-27-18_2e72ae25
      v2:
        snapshot: 2026-03-17T04-27-19_dd815223
      v3:
        snapshot: 2026-03-17T04-27-20_0592cbb0
      v4:
        snapshot: 2026-03-17T04-27-21_c2f019fb
      v5:
        snapshot: 2026-03-17T04-27-23_8ef8fcc4
    room_allocation_features:
      v1:
        snapshot: 2026-03-17T04-27-28_0368fd56
      v2:
        snapshot: 2026-03-17T04-27-29_e07095bc

Required snapshots are selected based on snapshot id at runtime. Extra dataset and feature set snapshots do not break the execution. Missing ones do.

3. Fields

Dataset Snapshot Bindings

Dataset snapshot bindings follow the following nesting policy:

datasets:
  └─ {dataset_name}
       └─ {dataset_version}
            └─ snapshot_information

Each nested layer specifies which snapshot should be used.

Example:

datasets:
hotel_bookings:
    v1:
    snapshot: 2026-03-17T03-26-25_4e817323
    v2:
    snapshot: 2026-03-17T03-26-27_c04ca84b

Feature Set Snapshot Bindings

Feature set snapshot bindings follow the following nesting policy:

feature_sets:
  └─ {feature_set_name}
       └─ {feature_set_version}
            └─ snapshot_information

Each nested layer specifies which snapshot should be used.

Example:

feature_sets:
    booking_context_features:
      v1:
        snapshot: 2026-03-17T06-41-04_530962ff
      v2:
        snapshot: 2026-03-17T04-27-05_b9099899
      v3:
        snapshot: 2026-03-17T04-27-07_6ecc2061
      v4:
        snapshot: 2026-03-17T04-27-08_e45a7462
      v5:
        snapshot: 2026-03-17T04-27-09_72290e7c
      v6:
        snapshot: 2026-03-17T04-27-10_f299b9d2
      v7:
        snapshot: 2026-03-17T04-27-12_eba7203e
    channel_and_agent_features:
      v1:
        snapshot: 2026-03-17T04-27-24_d2b1a0df
      v2:
        snapshot: 2026-03-17T04-27-25_1bd99237
      v3:
        snapshot: 2026-03-17T04-27-27_e73d2e8c

4. Validation

Snapshot bindings are validated using the validate_snapshot_binding_registry and validate_snapshot_binding functions.

Validation ensures:

  • the config file is formatted properly
  • snapshot is not empty
  • snapshot contains dataset or feature set bindings, if required

Invalid configurations raise a ConfigError.