Module encode

Source

Expand description

Observation/action encodings and schema constants. Observation/action encoding and spec helpers.

Related docs:

Structs§

ActionFactorizationSpec: Factorization schema for the action space.
ActionFamilySpec: Action family specification.
ActionIdDesc: Human-readable description of an action id.
ActionParam: Named parameter for an action id description.
ActionSpec: Full action specification.
FactorizedActionDesc: Machine-friendly factorized description of an action id.
ObsFieldSpec: Single observation field specification.
ObsSliceSpec: Slice specification for contiguous observation segments.
ObservationSpec: Full observation specification.
PlayerBlockSpec: Per-player observation block specification.

Enums§

ActionParamValue: Parameter value for an action id description.

Constants§

ACTION_ENCODING_VERSION: Action encoding version. Changelog: https://github.com/victorwp288/weiss-schwarz-simulator/blob/main/docs/encodings_changelog.md#action_encoding_version-1
ACTION_META_UNUSED: Sentinel value used for unused action metadata arguments.
ACTION_META_WIDTH: Number of u16 fields exported for each legal action metadata row.
ACTION_SPACE_SIZE: Total action space size.
ACTION_SPACE_WORDS: Number of u64 words required to represent the action mask.
ACTOR_NONE: Sentinel for missing actor.
ATTACK_BASE: Base action id for “attack” (slot and attack-type offsets).
ATTACK_COUNT: Number of attack actions.
ATTACK_SLOT_COUNT: Number of attack slots encoded/considered (front row).
CHOICE_BASE: Base action id for “choice select” (index offset).
CHOICE_COUNT: Number of choice select actions per page.
CHOICE_NEXT_ID: Action id for “next choice page”.
CHOICE_PREV_ID: Action id for “previous choice page”.
CLIMAX_PLAY_BASE: Base action id for “climax: play climax” (hand index offset).
CLIMAX_PLAY_COUNT: Number of climax play actions.
CLOCK_HAND_BASE: Base action id for “clock from hand” (hand index offset).
CLOCK_HAND_COUNT: Number of clock-from-hand actions (one per possible hand index).
CONCEDE_ID: Action id for “concede”.
DECISION_KIND_NONE: Sentinel for no decision.
ENCORE_DECLINE_BASE: Base action id for “encore: decline” (slot offset).
ENCORE_DECLINE_COUNT: Number of encore decline actions.
ENCORE_PAY_BASE: Base action id for “encore: pay” (slot offset).
ENCORE_PAY_COUNT: Number of encore pay actions.
LEVEL_UP_BASE: Base action id for “level up” (index offset).
LEVEL_UP_COUNT: Number of level-up actions.
MAIN_MOVE_BASE: Base action id for “main: move” (from/to slot offsets).
MAIN_MOVE_COUNT: Number of main move actions.
MAIN_PLAY_CHAR_BASE: Base action id for “main: play character” (hand index and slot offsets).
MAIN_PLAY_CHAR_COUNT: Number of main play character actions.
MAIN_PLAY_EVENT_BASE: Base action id for “main: play event” (hand index offset).
MAIN_PLAY_EVENT_COUNT: Number of main play event actions.
MAX_ABILITIES_PER_CARD: Maximum number of abilities encoded per card (padding beyond this is ignored).
MAX_DECK: Deck size per player.
MAX_HAND: Maximum hand size encoded in observations.
MAX_LEVEL: Maximum number of cards encoded in the level zone.
MAX_STAGE: Number of stage slots per player.
MULLIGAN_CONFIRM_ID: Action id for “confirm mulligan”.
MULLIGAN_SELECT_BASE: Base action id for “mulligan select” (hand index offset).
MULLIGAN_SELECT_COUNT: Number of mulligan select actions (one per possible hand index).
OBS_CONTEXT_BASE: Base index of the context slice.
OBS_CONTEXT_CHOICE_ACTIVE: Context bit/index: choice selection active.
OBS_CONTEXT_ENCORE_PENDING: Context bit/index: encore is pending.
OBS_CONTEXT_LEN: Length of the “context” slice.
OBS_CONTEXT_PRIORITY_WINDOW: Context bit/index: priority window active.
OBS_CONTEXT_STACK_NONEMPTY: Context bit/index: stack is non-empty.
OBS_ENCODING_VERSION: Observation encoding version. Changelog: https://github.com/victorwp288/weiss-schwarz-simulator/blob/main/docs/encodings_changelog.md#obs_encoding_version-2
OBS_HEADER_LEN: Observation header length.
OBS_LEN: Total observation vector length.
OBS_REASON_BASE: Base index of the observation reason slice.
OBS_REASON_IN_ATTACK: Reason bit/index: in attack phase.
OBS_REASON_IN_CLIMAX: Reason bit/index: in climax phase.
OBS_REASON_IN_COUNTER_WINDOW: Reason bit/index: in counter window.
OBS_REASON_IN_MAIN: Reason bit/index: in main phase.
OBS_REASON_LEN: Length of the per-observation “reason” slice.
OBS_REASON_NO_COLOR: Reason bit/index: missing color requirement.
OBS_REASON_NO_HAND: Reason bit/index: no cards in hand.
OBS_REASON_NO_STOCK: Reason bit/index: no stock available.
OBS_REASON_NO_TARGETS: Reason bit/index: no valid targets.
OBS_REVEAL_BASE: Base index of the reveal history slice.
OBS_REVEAL_LEN: Length of the “reveal history” slice.
PASS_ACTION_ID: Action id for “pass”.
PER_PLAYER_BLOCK_LEN: Per-player observation block length.
PER_PLAYER_CLIMAX_TOP: Cards encoded from the climax zone top.
PER_PLAYER_CLOCK_TOP: Cards encoded from the clock zone top.
PER_PLAYER_COUNTS: Number of scalar count slots per player block.
PER_PLAYER_DECK: Cards encoded from the deck.
PER_PLAYER_HAND: Cards encoded from the hand.
PER_PLAYER_LEVEL: Cards encoded from the level zone.
PER_PLAYER_RESOLUTION_TOP: Cards encoded from the resolution top.
PER_PLAYER_STAGE: Total per-player stage slice length.
PER_PLAYER_STOCK_TOP: Cards encoded from the stock top.
PER_PLAYER_WAITING_TOP: Cards encoded from the waiting room top.
PER_STAGE_SLOT: Scalars encoded per stage slot.
POLICY_VERSION: Policy version used in spec hash composition.
SPEC_HASH: Combined encoding spec hash.
TOP_CLOCK: Number of top cards encoded from the clock zone.
TOP_RESOLUTION: Number of top cards encoded from the resolution zone.
TOP_STOCK: Number of top cards encoded from the stock zone.
TOP_WAITING_ROOM: Number of top cards encoded from the waiting room zone.
TRIGGER_ORDER_BASE: Base action id for “trigger order” (index offset).
TRIGGER_ORDER_COUNT: Number of trigger-order actions.

Functions§

action_desc_for_id: Decode an action id into a canonical action descriptor.
action_id_for: Encode a canonical action descriptor into an action id.
action_spec: Build the action specification.
action_spec_json: Serialize the action specification as JSON.
build_action_mask: Build a dense action mask and lookup table from a legal action list.
decode_action_id: Decode an action id into a human-readable description.
decode_factorized_action_id: Decode an action id into a factorized family/argument description.
encode_factorized_action: Encode a factorized family/argument description into an action id.
encode_observation: Encode a full observation into a fixed-length buffer.
fill_action_mask: Fill a dense action mask and lookup table from a legal action list.
fill_action_mask_sparse: Update sparse action mask buffers from a legal action list.
observation_spec: Build the observation specification.
observation_spec_json: Serialize the observation specification as JSON.

Module encode

Module encode Copy item path

Structs§

Enums§

Constants§

Functions§

Module encode