rasa-pro 3.12.0.dev1__py3-none-any.whl

rasa/dialogue_understanding/processor/command_processor.py

@@ -0,0 +1,720 @@
+from typing import List, Optional, Type, Set, Dict
+
+import structlog
+from rasa.shared.core.training_data.structures import StoryGraph
+from rasa.dialogue_understanding.commands import (
+    CancelFlowCommand,
+    ClarifyCommand,
+    Command,
+    CorrectSlotsCommand,
+    CorrectedSlot,
+    RepeatBotMessagesCommand,
+    SetSlotCommand,
+    StartFlowCommand,
+    FreeFormAnswerCommand,
+    ChitChatAnswerCommand,
+    CannotHandleCommand,
+)
+from rasa.dialogue_understanding.commands.handle_code_change_command import (
+    HandleCodeChangeCommand,
+)
+from rasa.dialogue_understanding.commands.set_slot_command import SetSlotExtractor
+from rasa.dialogue_understanding.patterns.chitchat import FLOW_PATTERN_CHITCHAT
+from rasa.dialogue_understanding.patterns.collect_information import (
+    CollectInformationPatternFlowStackFrame,
+)
+from rasa.dialogue_understanding.patterns.correction import (
+    CorrectionPatternFlowStackFrame,
+)
+from rasa.dialogue_understanding.stack.dialogue_stack import DialogueStack
+from rasa.dialogue_understanding.stack.frames import (
+    BaseFlowStackFrame,
+)
+from rasa.dialogue_understanding.stack.utils import (
+    filled_slots_for_active_flow,
+    top_flow_frame,
+)
+from rasa.engine.graph import ExecutionContext
+from rasa.shared.constants import (
+    ROUTE_TO_CALM_SLOT,
+    RASA_PATTERN_CANNOT_HANDLE_CHITCHAT,
+)
+from rasa.shared.core.constants import ACTION_TRIGGER_CHITCHAT, SlotMappingType
+from rasa.shared.core.constants import FLOW_HASHES_SLOT
+from rasa.shared.core.events import Event, SlotSet
+from rasa.shared.core.flows import FlowsList
+from rasa.shared.core.flows.steps.collect import CollectInformationFlowStep
+from rasa.shared.core.slots import Slot
+from rasa.shared.core.trackers import DialogueStateTracker
+from rasa.shared.nlu.constants import COMMANDS
+
+structlogger = structlog.get_logger()
+
+CANNOT_HANDLE_REASON = (
+    "A command generator attempted to set a slot "
+    "with a value extracted by an extractor "
+    "that is incompatible with the slot mapping type."
+)
+
+
+def contains_command(commands: List[Command], typ: Type[Command]) -> bool:
+    """Check if a list of commands contains a command of a given type.
+
+    Example:
+        >>> contains_command([SetSlotCommand("foo", "bar")], SetSlotCommand)
+        True
+
+    Args:
+        commands: The commands to check.
+        typ: The type of command to check for.
+
+    Returns:
+    `True` if the list of commands contains a command of the given type.
+    """
+    return any(isinstance(command, typ) for command in commands)
+
+
+def get_commands_from_tracker(tracker: DialogueStateTracker) -> List[Command]:
+    """Extracts the commands from the tracker.
+
+    Args:
+        tracker: The tracker containing the conversation history up to now.
+
+    Returns:
+    The commands.
+    """
+    if tracker.latest_message:
+        dumped_commands = tracker.latest_message.parse_data.get(COMMANDS) or []
+        assert isinstance(dumped_commands, list)
+        return [Command.command_from_json(command) for command in dumped_commands]
+    else:
+        return []
+
+
+def filter_start_flow_commands(commands: List[Command]) -> List[str]:
+    """Filters the start flow commands from a list of commands."""
+    return [
+        command.flow for command in commands if isinstance(command, StartFlowCommand)
+    ]
+
+
+def validate_state_of_commands(commands: List[Command]) -> None:
+    """Validates the state of the commands.
+
+    We have some invariants that should always hold true. This function
+    checks if they do. Executing the commands relies on these invariants.
+
+    We cleanup the commands before executing them, so the cleanup should
+    always make sure that these invariants hold true - no matter the commands
+    that are provided.
+
+    Args:
+        commands: The commands to validate.
+
+    Raises:
+        ValueError: If the state of the commands is invalid.
+    """
+    # check that there is only at max one cancel flow command
+    if sum(isinstance(c, CancelFlowCommand) for c in commands) > 1:
+        structlogger.error(
+            "command_processor.validate_state_of_commands.multiple_cancel_flow_commands",
+            commands=commands,
+        )
+        raise ValueError("There can only be one cancel flow command.")
+
+    # check that free form answer commands are only at the beginning of the list
+    free_form_answer_commands = [
+        c for c in commands if isinstance(c, FreeFormAnswerCommand)
+    ]
+    if free_form_answer_commands != commands[: len(free_form_answer_commands)]:
+        structlogger.error(
+            "command_processor.validate_state_of_commands.free_form_answer_commands_not_at_beginning",
+            commands=commands,
+        )
+        raise ValueError(
+            "Free form answer commands must be at start of the predicted command list."
+        )
+
+    # check that there is at max only one correctslots command
+    if sum(isinstance(c, CorrectSlotsCommand) for c in commands) > 1:
+        structlogger.error(
+            "command_processor.validate_state_of_commands.multiple_correct_slots_commands",
+            commands=commands,
+        )
+        raise ValueError("There can only be one correct slots command.")
+
+
+def find_updated_flows(tracker: DialogueStateTracker, all_flows: FlowsList) -> Set[str]:
+    """Find the set of updated flows.
+
+    Run through the current dialogue stack and compare the flow hashes of the
+    flows on the stack with those stored in the tracker.
+
+    Args:
+        tracker: The tracker.
+        all_flows: All flows.
+
+    Returns:
+    A set of flow ids of those flows that have changed
+    """
+    stored_fingerprints: Dict[str, str] = tracker.get_slot(FLOW_HASHES_SLOT) or {}
+    stack = tracker.stack
+
+    changed_flows = set()
+    for frame in stack.frames:
+        if isinstance(frame, BaseFlowStackFrame):
+            flow = all_flows.flow_by_id(frame.flow_id)
+            if flow is None or (
+                flow.id in stored_fingerprints
+                and flow.fingerprint != stored_fingerprints[flow.id]
+            ):
+                changed_flows.add(frame.flow_id)
+    return changed_flows
+
+
+def calculate_flow_fingerprints(all_flows: FlowsList) -> Dict[str, str]:
+    """Calculate fingerprints for all flows."""
+    return {flow.id: flow.fingerprint for flow in all_flows.underlying_flows}
+
+
+def execute_commands(
+    tracker: DialogueStateTracker,
+    all_flows: FlowsList,
+    execution_context: ExecutionContext,
+    story_graph: Optional[StoryGraph] = None,
+) -> List[Event]:
+    """Executes a list of commands.
+
+    Args:
+        commands: The commands to execute.
+        tracker: The tracker to execute the commands on.
+        all_flows: All flows.
+        execution_context: Information about the single graph run.
+        story_graph: StoryGraph object with stories available for training.
+
+    Returns:
+        A list of the events that were created.
+    """
+    commands: List[Command] = get_commands_from_tracker(tracker)
+    original_tracker = tracker.copy()
+
+    commands = clean_up_commands(
+        commands, tracker, all_flows, execution_context, story_graph
+    )
+
+    updated_flows = find_updated_flows(tracker, all_flows)
+    if updated_flows:
+        # Override commands
+        structlogger.debug(
+            "command_processor.execute_commands.running_flows_were_updated",
+            updated_flow_ids=updated_flows,
+        )
+        commands = [HandleCodeChangeCommand()]
+
+    # store current flow hashes if they changed
+    new_hashes = calculate_flow_fingerprints(all_flows)
+    flow_hash_events: List[Event] = []
+    if new_hashes != (tracker.get_slot(FLOW_HASHES_SLOT) or {}):
+        flow_hash_events.append(SlotSet(FLOW_HASHES_SLOT, new_hashes))
+        tracker.update_with_events(flow_hash_events)
+
+    events: List[Event] = flow_hash_events
+
+    # commands need to be reversed to make sure they end up in the right order
+    # on the stack. e.g. if there multiple start flow commands, the first one
+    # should be on top of the stack. this is achieved by reversing the list
+    # and then pushing the commands onto the stack in the reversed order.
+    reversed_commands = list(reversed(commands))
+
+    validate_state_of_commands(commands)
+
+    for command in reversed_commands:
+        new_events = command.run_command_on_tracker(
+            tracker, all_flows, original_tracker
+        )
+        events.extend(new_events)
+        tracker.update_with_events(new_events)
+
+    return remove_duplicated_set_slots(events)
+
+
+def remove_duplicated_set_slots(events: List[Event]) -> List[Event]:
+    """Removes duplicated set slot events.
+
+    This can happen if a slot is set multiple times in a row. We only want to
+    keep the last one.
+
+    Args:
+        events: The events to optimize.
+
+    Returns:
+    The optimized events.
+    """
+    slots_so_far = set()
+
+    optimized_events: List[Event] = []
+
+    for event in reversed(events):
+        if isinstance(event, SlotSet) and event.key in slots_so_far:
+            # slot will be overwritten, no need to set it
+            continue
+        elif isinstance(event, SlotSet):
+            slots_so_far.add(event.key)
+
+        optimized_events.append(event)
+
+    # since we reversed the original events, we need to reverse the optimized
+    # events again to get them in the right order
+    return list(reversed(optimized_events))
+
+
+def get_current_collect_step(
+    dialogue_stack: DialogueStack, all_flows: FlowsList
+) -> Optional[CollectInformationFlowStep]:
+    """Get the current collect information if the conversation is currently in one.
+
+    If we are currently in a collect information step, the stack should have at least
+    two frames. The top frame is the collect information pattern and the frame below
+    is the flow that triggered the collect information pattern. We can use the flow
+    id to get the collect information step from the flow.
+
+    Args:
+        dialogue_stack: The dialogue stack.
+        all_flows: All flows.
+
+    Returns:
+    The current collect information if the conversation is currently in one,
+    `None` otherwise.
+    """
+    if not (top_frame := dialogue_stack.top()):
+        # we are currently not in a flow
+        return None
+
+    if not isinstance(top_frame, CollectInformationPatternFlowStackFrame):
+        # we are currently not in a collect information
+        return None
+
+    if len(dialogue_stack.frames) <= 1:
+        # for some reason only the collect information pattern step is on the stack
+        # but no flow that triggered it. this should never happen.
+        structlogger.warning(
+            "command_processor.get_current_collect_step.no_flow_on_stack",
+            stack=dialogue_stack,
+        )
+        return None
+
+    frame_that_triggered_collect_infos = dialogue_stack.frames[-2]
+    if not isinstance(frame_that_triggered_collect_infos, BaseFlowStackFrame):
+        # this is a failure, if there is a frame, we should be able to get the
+        # step from it
+        structlogger.warning(
+            "command_processor.get_current_collect_step.no_step_for_frame",
+            frame=frame_that_triggered_collect_infos,
+        )
+        return None
+
+    step = frame_that_triggered_collect_infos.step(all_flows)
+    if isinstance(step, CollectInformationFlowStep):
+        # we found it!
+        return step
+    else:
+        # this should never happen as we only push collect information patterns
+        # onto the stack if there is a collect information step
+        structlogger.warning(
+            "command_processor.get_current_collect_step.step_not_collect",
+            step=step,
+        )
+        return None
+
+
+def clean_up_commands(
+    commands: List[Command],
+    tracker: DialogueStateTracker,
+    all_flows: FlowsList,
+    execution_context: ExecutionContext,
+    story_graph: Optional[StoryGraph] = None,
+) -> List[Command]:
+    """Clean up a list of commands.
+
+    This will clean commands that are not necessary anymore. e.g. removing commands
+    where the slot they correct was previously corrected to the same value, grouping
+    all slot corrections into one command, removing duplicate cancel flow commands
+    and moving free form answer commands to the beginning of the list (to be last when
+    reversed.)
+
+    Args:
+        commands: The commands to clean up.
+        tracker: The tracker to clean up the commands for.
+        all_flows: All flows.
+        execution_context: Information about a single graph run.
+        story_graph: StoryGraph object with stories available for training.
+
+    Returns:
+    The cleaned up commands.
+    """
+    slots_so_far, active_flow = filled_slots_for_active_flow(tracker, all_flows)
+
+    clean_commands: List[Command] = []
+
+    for command in commands:
+        if isinstance(command, SetSlotCommand):
+            clean_commands = clean_up_slot_command(
+                clean_commands, command, tracker, all_flows, slots_so_far
+            )
+
+        elif isinstance(command, CancelFlowCommand) and contains_command(
+            clean_commands, CancelFlowCommand
+        ):
+            structlogger.debug(
+                "command_processor.clean_up_commands"
+                ".skip_command_flow_already_cancelled",
+                command=command,
+            )
+
+        # if there is a cannot handle command after the previous step,
+        # we don't want to add another one
+        elif isinstance(command, CannotHandleCommand) and contains_command(
+            clean_commands, CannotHandleCommand
+        ):
+            structlogger.debug(
+                "command_processor"
+                ".clean_up_commands"
+                ".skip_command_already_has_cannot_handle",
+                command=command,
+            )
+
+        elif isinstance(command, StartFlowCommand) and command.flow == active_flow:
+            # drop a start flow command if the starting flow is equal to the currently
+            # active flow
+            structlogger.debug(
+                "command_processor.clean_up_commands.skip_command_flow_already_active",
+                command=command,
+            )
+
+        # handle chitchat command differently from other free-form answer commands
+        elif isinstance(command, ChitChatAnswerCommand):
+            clean_commands = clean_up_chitchat_command(
+                clean_commands, command, all_flows, execution_context, story_graph
+            )
+
+        elif isinstance(command, FreeFormAnswerCommand):
+            structlogger.debug(
+                "command_processor.clean_up_commands.prepend_command_freeform_answer",
+                command=command,
+            )
+            clean_commands.insert(0, command)
+
+        # drop all clarify commands if there are more commands. Otherwise, we might
+        # get a situation where two questions are asked at once.
+        elif isinstance(command, ClarifyCommand) and len(commands) > 1:
+            clean_commands = clean_up_clarify_command(clean_commands, commands, command)
+            if command not in clean_commands:
+                structlogger.debug(
+                    "command_processor.clean_up_commands.drop_clarify_given_other_commands",
+                    command=command,
+                )
+        else:
+            clean_commands.append(command)
+
+    # when coexistence is enabled, by default there will be a SetSlotCommand
+    # for the ROUTE_TO_CALM_SLOT slot.
+    if tracker.has_coexistence_routing_slot and len(clean_commands) > 2:
+        clean_commands = filter_cannot_handle_command_for_skipped_slots(clean_commands)
+    elif not tracker.has_coexistence_routing_slot and len(clean_commands) > 1:
+        clean_commands = filter_cannot_handle_command_for_skipped_slots(clean_commands)
+
+    clean_commands = ensure_max_number_of_command_type(
+        clean_commands, RepeatBotMessagesCommand, 1
+    )
+    structlogger.debug(
+        "command_processor.clean_up_commands.final_commands",
+        command=clean_commands,
+    )
+
+    return clean_commands
+
+
+def ensure_max_number_of_command_type(
+    commands: List[Command], command_type: Type[Command], n: int
+) -> List[Command]:
+    """Ensures that for a given command type only the first n stay in the list."""
+    filtered: List[Command] = []
+    count = 0
+    for c in commands:
+        if isinstance(c, command_type):
+            if count >= n:
+                continue
+            else:
+                count += 1
+        filtered.append(c)
+    return filtered
+
+
+def clean_up_clarify_command(
+    commands_so_far: List[Command],
+    all_commands: List[Command],
+    current_command: ClarifyCommand,
+) -> List[Command]:
+    """Clean up a clarify command.
+
+    Args:
+        commands_so_far: The commands cleaned up so far.
+        all_commands: All the predicted commands.
+        current_command: The current clarify command.
+
+    Returns:
+        The cleaned up commands.
+    """
+    # Get the commands after removing the ROUTE_TO_CALM_SLOT set slot command.
+    commands_without_route_to_calm_set_slot = [
+        c
+        for c in all_commands
+        if not (isinstance(c, SetSlotCommand) and c.name == ROUTE_TO_CALM_SLOT)
+    ]
+
+    # if all commands are clarify commands, add the first one only, otherwise add none
+    if all(
+        isinstance(c, ClarifyCommand) for c in commands_without_route_to_calm_set_slot
+    ):
+        # Check if clean_commands is empty or contains only ROUTE_TO_CALM_SLOT
+        # set slot command.
+        if not commands_so_far or (
+            len(commands_so_far) == 1
+            and isinstance(commands_so_far[0], SetSlotCommand)
+            and commands_so_far[0].name == ROUTE_TO_CALM_SLOT
+        ):
+            commands_so_far.append(current_command)
+
+    return commands_so_far
+
+
+def clean_up_slot_command(
+    commands_so_far: List[Command],
+    command: SetSlotCommand,
+    tracker: DialogueStateTracker,
+    all_flows: FlowsList,
+    slots_so_far: Set[str],
+) -> List[Command]:
+    """Clean up a slot command.
+
+    This will remove commands that are not necessary anymore, e.g. because the slot
+    they correct was previously corrected to the same value. It will group all slot
+    corrections into one command.
+
+    Args:
+        commands_so_far: The commands cleaned up so far.
+        command: The command to clean up.
+        tracker: The dialogue state tracker.
+        all_flows: All flows.
+        slots_so_far: The slots that have been filled so far.
+
+    Returns:
+        The cleaned up commands.
+    """
+    stack = tracker.stack
+
+    resulting_commands = commands_so_far[:]
+
+    slot = tracker.slots.get(command.name)
+    if slot is None:
+        structlogger.debug(
+            "command_processor.clean_up_slot_command.skip_command_slot_not_in_domain",
+            command=command,
+        )
+        return resulting_commands
+
+    if not should_slot_be_set(slot, command):
+        cannot_handle = CannotHandleCommand(reason=CANNOT_HANDLE_REASON)
+        if cannot_handle not in resulting_commands:
+            resulting_commands.append(cannot_handle)
+
+        return resulting_commands
+
+    if command.name in slots_so_far and command.name != ROUTE_TO_CALM_SLOT:
+        current_collect_info = get_current_collect_step(stack, all_flows)
+
+        if current_collect_info and current_collect_info.collect == command.name:
+            # not a correction but rather an answer to the current collect info
+            resulting_commands.append(command)
+            return resulting_commands
+
+        if (slot := tracker.slots.get(command.name)) is not None and slot.value == str(
+            command.value
+        ):
+            # the slot is already set, we don't need to set it again
+            structlogger.debug(
+                "command_processor.clean_up_slot_command.skip_command_slot_already_set",
+                command=command,
+            )
+            return resulting_commands
+
+        top = top_flow_frame(stack)
+        if isinstance(top, CorrectionPatternFlowStackFrame):
+            already_corrected_slots = top.corrected_slots
+        else:
+            already_corrected_slots = {}
+
+        if command.name in already_corrected_slots and str(
+            already_corrected_slots[command.name]
+        ) == str(command.value):
+            structlogger.debug(
+                "command_processor.clean_up_slot_command"
+                ".skip_command_slot_already_corrected",
+                command=command,
+            )
+            return resulting_commands
+
+        structlogger.debug(
+            "command_processor.clean_up_slot_command.convert_command_to_correction",
+            command=command,
+        )
+
+        # Group all corrections into one command
+        corrected_slot = CorrectedSlot(command.name, command.value)
+        for c in resulting_commands:
+            if isinstance(c, CorrectSlotsCommand):
+                c.corrected_slots.append(corrected_slot)
+                break
+        else:
+            resulting_commands.append(
+                CorrectSlotsCommand(corrected_slots=[corrected_slot])
+            )
+    else:
+        resulting_commands.append(command)
+
+    return resulting_commands
+
+
+def clean_up_chitchat_command(
+    commands_so_far: List[Command],
+    command: ChitChatAnswerCommand,
+    flows: FlowsList,
+    execution_context: ExecutionContext,
+    story_graph: Optional[StoryGraph] = None,
+) -> List[Command]:
+    """Clean up a chitchat answer command.
+
+    Respond with 'cannot handle' if 'IntentlessPolicy' is unset in
+    model config but 'action_trigger_chitchat' is used within the pattern_chitchat
+
+    Args:
+        commands_so_far: The commands cleaned up so far.
+        command: The command to clean up.
+        flows: All flows.
+        execution_context: Information about a single graph run.
+        story_graph: StoryGraph object with stories available for training.
+    Returns:
+        The cleaned up commands.
+    """
+    from rasa.core.policies.intentless_policy import IntentlessPolicy
+
+    resulting_commands = commands_so_far[:]
+
+    pattern_chitchat = flows.flow_by_id(FLOW_PATTERN_CHITCHAT)
+
+    # very unlikely to happen, placed here due to mypy checks
+    if pattern_chitchat is None:
+        resulting_commands.insert(
+            0, CannotHandleCommand(RASA_PATTERN_CANNOT_HANDLE_CHITCHAT)
+        )
+        structlogger.warn(
+            "command_processor.clean_up_chitchat_command.pattern_chitchat_not_found",
+            command=resulting_commands[0],
+        )
+        return resulting_commands
+
+    has_action_trigger_chitchat = pattern_chitchat.has_action_step(
+        ACTION_TRIGGER_CHITCHAT
+    )
+    defines_intentless_policy = execution_context.has_node(IntentlessPolicy)
+
+    has_e2e_stories = True if (story_graph and story_graph.has_e2e_stories()) else False
+
+    if (has_action_trigger_chitchat and not defines_intentless_policy) or (
+        defines_intentless_policy and not has_e2e_stories
+    ):
+        resulting_commands.insert(
+            0, CannotHandleCommand(RASA_PATTERN_CANNOT_HANDLE_CHITCHAT)
+        )
+        structlogger.warn(
+            "command_processor.clean_up_chitchat_command.replace_chitchat_answer_with_cannot_handle",
+            command=resulting_commands[0],
+            pattern_chitchat_uses_action_trigger_chitchat=has_action_trigger_chitchat,
+            defined_intentless_policy_in_config=defines_intentless_policy,
+        )
+    else:
+        resulting_commands.insert(0, command)
+        structlogger.debug(
+            "command_processor.clean_up_commands.prepend_command_chitchat_answer",
+            command=command,
+            pattern_chitchat_uses_action_trigger_chitchat=has_action_trigger_chitchat,
+            defined_intentless_policy_in_config=defines_intentless_policy,
+        )
+
+    return resulting_commands
+
+
+def should_slot_be_set(slot: Slot, command: SetSlotCommand) -> bool:
+    """Check if a slot should be set by a command."""
+    if command.extractor == SetSlotExtractor.COMMAND_PAYLOAD_READER.value:
+        # if the command is issued by the command payload reader, it means the slot
+        # was set deterministically via a response button. In this case,
+        # we can always set it
+        return True
+
+    slot_mappings = slot.mappings
+
+    if not slot_mappings:
+        slot_mappings = [{"type": SlotMappingType.FROM_LLM.value}]
+
+    for mapping in slot_mappings:
+        mapping_type = SlotMappingType(
+            mapping.get("type", SlotMappingType.FROM_LLM.value)
+        )
+
+        should_be_set_by_llm = (
+            command.extractor == SetSlotExtractor.LLM.value
+            and mapping_type == SlotMappingType.FROM_LLM
+        )
+        should_be_set_by_nlu = (
+            command.extractor == SetSlotExtractor.NLU.value
+            and mapping_type.is_predefined_type()
+        )
+
+        if should_be_set_by_llm or should_be_set_by_nlu:
+            # if the extractor matches the mapping type, we can continue
+            # setting the slot
+            break
+
+        structlogger.debug(
+            "command_processor.clean_up_slot_command.skip_command.extractor_"
+            "does_not_match_slot_mapping",
+            extractor=command.extractor,
+            slot_name=slot.name,
+            mapping_type=mapping_type.value,
+        )
+        return False
+
+    return True
+
+
+def filter_cannot_handle_command_for_skipped_slots(
+    clean_commands: List[Command],
+) -> List[Command]:
+    """Filter out a 'cannot handle' command for skipped slots.
+
+    This is used to filter out a 'cannot handle' command for skipped slots
+    in case other commands are present.
+
+    Returns:
+        The filtered commands.
+    """
+    return [
+        command
+        for command in clean_commands
+        if not (
+            isinstance(command, CannotHandleCommand)
+            and command.reason
+            and CANNOT_HANDLE_REASON == command.reason
+        )
+    ]