Upd/dff/template skill (#303)

* add dff_release folder; add minimal changes to server.py; modify dockerfile to use dff_release; modify tests for use with message class; ensure the passing of tests * rename actor.py to serializer.py, reimplement serialization * edit requirements.txt; use MultiMessage from dff * rename dff_release to dff_api_v1; change dockerfile to import dff_api_v1; remove nulls from skill output; exclude nulls from test data; add run_dff function to serializer file * add DreamMessage and DreamMultiMessage classes; process in serializer; make message fields optional to discard null values; edit main.py * remove redundant imports from serializer.py && main.py * remove DreamMultiMessage * Move Pre_transition_processing from node_1 to start_node * apply lint * Add comments to main.py; remove actor instance * apply lint * update requirements: define versions; update import order in main * do minor fixes * move slot filling to a separate integration.response module * change function signature. successful building of dream skill * fix codestyle * merge upstream dev: * Use new dff version in the requirements; --------- Co-authored-by: dilyararimovna <[email protected]>
deeppavlov · Jun 19, 2023 · 04dbfbb · 04dbfbb
1 parent 026ca9e
commit 04dbfbb
Show file tree

Hide file tree

Showing 15 changed files with 1,354 additions and 94 deletions.
diff --git a/common/dff_api_v1/integration/condition.py b/common/dff_api_v1/integration/condition.py
@@ -0,0 +1,289 @@
+import logging
+import re
+
+from nltk.stem import WordNetLemmatizer
+
+from dff.script import Context
+from dff.pipeline import Pipeline
+
+import common.greeting as common_greeting
+import common.utils as common_utils
+import common.universal_templates as universal_templates
+import common.dff_api_v1.integration.context as int_ctx
+from common.acknowledgements import GENERAL_ACKNOWLEDGEMENTS
+from common.constants import CAN_CONTINUE_SCENARIO, CAN_NOT_CONTINUE
+from .facts_utils import provide_facts_request
+
+logger = logging.getLogger(__name__)
+
+wnl = WordNetLemmatizer()
+
+
+#  vars is described in README.md
+
+
+def was_clarification_request(ctx: Context, _) -> bool:
+    flag = ctx.misc["agent"]["clarification_request_flag"] if not ctx.validation else False
+    logger.debug(f"was_clarification_request = {flag}")
+    return bool(flag)
+
+
+def is_opinion_request(ctx: Context, pipeline: Pipeline) -> bool:
+    flag = common_utils.is_opinion_request(int_ctx.get_last_human_utterance(ctx, pipeline))
+    logger.debug(f"is_opinion_request = {flag}")
+    return bool(flag)
+
+
+def is_opinion_expression(ctx: Context, pipeline: Pipeline) -> bool:
+    flag = common_utils.is_opinion_expression(int_ctx.get_last_human_utterance(ctx, pipeline))
+    logger.debug(f"is_opinion_expression = {flag}")
+    return bool(flag)
+
+
+def is_previous_turn_dff_suspended(ctx: Context, _) -> bool:
+    flag = ctx.misc["agent"].get("previous_turn_dff_suspended", False) if not ctx.validation else False
+    logger.debug(f"is_previous_turn_dff_suspended = {flag}")
+    return bool(flag)
+
+
+def is_current_turn_dff_suspended(ctx: Context, _) -> bool:
+    flag = ctx.misc["agent"].get("current_turn_dff_suspended", False) if not ctx.validation else False
+    logger.debug(f"is_current_turn_dff_suspended = {flag}")
+    return bool(flag)
+
+
+def is_switch_topic(ctx: Context, pipeline: Pipeline) -> bool:
+    flag = universal_templates.is_switch_topic(int_ctx.get_last_human_utterance(ctx, pipeline))
+    logger.debug(f"is_switch_topic = {flag}")
+    return bool(flag)
+
+
+def is_question(ctx: Context, pipeline: Pipeline) -> bool:
+    text = int_ctx.get_last_human_utterance(ctx, pipeline)["text"]
+    flag = common_utils.is_question(text)
+    logger.debug(f"is_question = {flag}")
+    return bool(flag)
+
+
+def is_lets_chat_about_topic_human_initiative(ctx: Context, pipeline: Pipeline) -> bool:
+    flag = universal_templates.if_chat_about_particular_topic(
+        int_ctx.get_last_human_utterance(ctx, pipeline), int_ctx.get_last_bot_utterance(ctx, pipeline)
+    )
+    logger.debug(f"is_lets_chat_about_topic_human_initiative = {flag}")
+    return bool(flag)
+
+
+def is_lets_chat_about_topic(ctx: Context, pipeline: Pipeline) -> bool:
+    flag = is_lets_chat_about_topic_human_initiative(ctx, pipeline)
+
+    last_human_uttr = int_ctx.get_last_human_utterance(ctx, pipeline)
+    last_bot_uttr_text = int_ctx.get_last_bot_utterance(ctx, pipeline)["text"]
+    is_bot_initiative = bool(re.search(universal_templates.COMPILE_WHAT_TO_TALK_ABOUT, last_bot_uttr_text))
+    flag = flag or (is_bot_initiative and not common_utils.is_no(last_human_uttr))
+    logger.debug(f"is_lets_chat_about_topic = {flag}")
+    return bool(flag)
+
+
+def is_begin_of_dialog(ctx: Context, pipeline: Pipeline, begin_dialog_n=10) -> bool:
+    flag = int_ctx.get_human_utter_index(ctx, pipeline) < begin_dialog_n
+    logger.debug(f"is_begin_of_dialog = {flag}")
+    return bool(flag)
+
+
+def is_interrupted(ctx: Context, pipeline: Pipeline) -> bool:
+    flag = (
+        int_ctx.get_human_utter_index(ctx, pipeline) - int_ctx.get_previous_human_utter_index(ctx, pipeline)
+    ) != 1 and not was_clarification_request(ctx, pipeline)
+    logger.debug(f"is_interrupted = {flag}")
+    return bool(flag)
+
+
+def is_long_interrupted(ctx: Context, pipeline: Pipeline, how_long=3) -> bool:
+    flag = (
+        int_ctx.get_human_utter_index(ctx, pipeline) - int_ctx.get_previous_human_utter_index(ctx, pipeline)
+    ) > how_long and not was_clarification_request(ctx, pipeline)
+    logger.debug(f"is_long_interrupted = {flag}")
+    return bool(flag)
+
+
+def is_new_human_entity(ctx: Context, pipeline: Pipeline) -> bool:
+    new_entities = int_ctx.get_new_human_labeled_noun_phrase(ctx, pipeline)
+    flag = bool(new_entities)
+    logger.debug(f"is_new_human_entity = {flag}")
+    return bool(flag)
+
+
+def is_last_state(ctx: Context, pipeline: Pipeline, state) -> bool:
+    flag = False
+    if not ctx.validation:
+        history = list(int_ctx.get_history(ctx, pipeline).items())
+        if history:
+            history_sorted = sorted(history, key=lambda x: x[0])
+            last_state = history_sorted[-1][1]
+            if last_state == state:
+                flag = True
+    return bool(flag)
+
+
+def is_first_time_of_state(ctx: Context, pipeline: Pipeline, state) -> bool:
+    flag = state not in list(int_ctx.get_history(ctx, pipeline).values())
+    logger.debug(f"is_first_time_of_state {state} = {flag}")
+    return bool(flag)
+
+
+def if_was_prev_active(ctx: Context, pipeline: Pipeline) -> bool:
+    flag = False
+    skill_uttr_indices = set(int_ctx.get_history(ctx, pipeline).keys())
+    if not ctx.validation:
+        human_uttr_index = str(ctx.misc["agent"]["human_utter_index"] - 1)
+        if human_uttr_index in skill_uttr_indices:
+            flag = True
+    return bool(flag)
+
+
+def is_plural(word) -> bool:
+    lemma = wnl.lemmatize(word, "n")
+    plural = True if word is not lemma else False
+    return plural
+
+
+def is_first_our_response(ctx: Context, pipeline: Pipeline) -> bool:
+    flag = len(list(int_ctx.get_history(ctx, pipeline).values())) == 0
+    logger.debug(f"is_first_our_response = {flag}")
+    return bool(flag)
+
+
+def is_no_human_abandon(ctx: Context, pipeline: Pipeline) -> bool:
+    """Is dialog breakdown in human utterance or no. Uses MIDAS hold/abandon classes."""
+    midas_classes = common_utils.get_intents(int_ctx.get_last_human_utterance(ctx, pipeline), which="midas")
+    if "abandon" not in midas_classes:
+        return True
+    return False
+
+
+def no_special_switch_off_requests(ctx: Context, pipeline: Pipeline) -> bool:
+    """Function to determine if
+    - user didn't asked to switch topic,
+    - user didn't ask to talk about something particular,
+    - user didn't requested high priority intents (like what_is_your_name)
+    """
+    intents_by_catcher = common_utils.get_intents(
+        int_ctx.get_last_human_utterance(ctx, pipeline), probs=False, which="intent_catcher"
+    )
+    is_high_priority_intent = any([intent not in common_utils.service_intents for intent in intents_by_catcher])
+    is_switch = is_switch_topic(ctx, pipeline)
+    is_lets_chat = is_lets_chat_about_topic_human_initiative(ctx, pipeline)
+
+    if not (is_high_priority_intent or is_switch or is_lets_chat):
+        return True
+    return False
+
+
+def no_requests(ctx: Context, pipeline: Pipeline) -> bool:
+    """Function to determine if
+    - user didn't asked to switch topic,
+    - user didn't ask to talk about something particular,
+    - user didn't requested high priority intents (like what_is_your_name)
+    - user didn't requested any special intents
+    - user didn't ask questions
+    """
+    contain_no_special_requests = no_special_switch_off_requests(ctx, pipeline)
+
+    request_intents = [
+        "opinion_request",
+        "topic_switching",
+        "lets_chat_about",
+        "what_are_you_talking_about",
+        "Information_RequestIntent",
+        "Topic_SwitchIntent",
+        "Opinion_RequestIntent",
+    ]
+    intents = common_utils.get_intents(int_ctx.get_last_human_utterance(ctx, pipeline), which="all")
+    is_not_request_intent = all([intent not in request_intents for intent in intents])
+    is_no_question = "?" not in int_ctx.get_last_human_utterance(ctx, pipeline)["text"]
+
+    if contain_no_special_requests and is_not_request_intent and is_no_question:
+        return True
+    return False
+
+
+def is_yes_vars(ctx: Context, pipeline: Pipeline) -> bool:
+    flag = True
+    flag = flag and common_utils.is_yes(int_ctx.get_last_human_utterance(ctx, pipeline))
+    return bool(flag)
+
+
+def is_no_vars(ctx: Context, pipeline: Pipeline) -> bool:
+    flag = True
+    flag = flag and common_utils.is_no(int_ctx.get_last_human_utterance(ctx, pipeline))
+    return bool(flag)
+
+
+def is_do_not_know_vars(ctx: Context, pipeline: Pipeline) -> bool:
+    flag = True
+    flag = flag and common_utils.is_donot_know(int_ctx.get_last_human_utterance(ctx, pipeline))
+    return bool(flag)
+
+
+def is_passive_user(ctx: Context, pipeline: Pipeline, passive_threshold=3, history_len=2) -> bool:
+    """Check history_len last human utterances on the number of tokens.
+    If number of tokens in ALL history_len uterances is less or equal than the given threshold,
+    then consider user passive - return True.
+    """
+    user_utterances = int_ctx.get_human_utterances(ctx, pipeline)[-history_len:]
+    user_utterances = [utt["text"] for utt in user_utterances]
+
+    uttrs_lens = [len(uttr.split()) <= passive_threshold for uttr in user_utterances]
+    if all(uttrs_lens):
+        return True
+    return False
+
+
+def get_not_used_and_save_sentiment_acknowledgement(ctx: Context, pipeline: Pipeline, sentiment=None, lang="EN"):
+    if sentiment is None:
+        sentiment = int_ctx.get_human_sentiment(ctx, pipeline)
+        if is_yes_vars(ctx, pipeline) or is_no_vars(ctx, pipeline):
+            sentiment = "neutral"
+
+    shared_memory = int_ctx.get_shared_memory(ctx, pipeline)
+    last_acknowledgements = shared_memory.get("last_acknowledgements", [])
+
+    ack = common_utils.get_not_used_template(
+        used_templates=last_acknowledgements, all_templates=GENERAL_ACKNOWLEDGEMENTS[lang][sentiment]
+    )
+
+    used_acks = last_acknowledgements + [ack]
+    int_ctx.save_to_shared_memory(ctx, pipeline, last_acknowledgements=used_acks[-2:])
+    return ack
+
+
+def set_conf_and_can_cont_by_universal_policy(ctx: Context, pipeline: Pipeline):
+    DIALOG_BEGINNING_START_CONFIDENCE = 0.98
+    DIALOG_BEGINNING_CONTINUE_CONFIDENCE = 0.9
+    DIALOG_BEGINNING_SHORT_ANSWER_CONFIDENCE = 0.98
+    MIDDLE_DIALOG_START_CONFIDENCE = 0.7
+
+    if not is_begin_of_dialog(ctx, pipeline, begin_dialog_n=10):
+        confidence = 0.0
+        can_continue_flag = CAN_NOT_CONTINUE
+    elif is_first_our_response(ctx, pipeline):
+        confidence = DIALOG_BEGINNING_START_CONFIDENCE
+        can_continue_flag = CAN_CONTINUE_SCENARIO
+    elif not is_interrupted(ctx, pipeline) and common_greeting.dont_tell_you_answer(
+        int_ctx.get_last_human_utterance(ctx, pipeline)
+    ):
+        confidence = DIALOG_BEGINNING_SHORT_ANSWER_CONFIDENCE
+        can_continue_flag = CAN_CONTINUE_SCENARIO
+    elif not is_interrupted(ctx, pipeline):
+        confidence = DIALOG_BEGINNING_CONTINUE_CONFIDENCE
+        can_continue_flag = CAN_CONTINUE_SCENARIO
+    else:
+        confidence = MIDDLE_DIALOG_START_CONFIDENCE
+        can_continue_flag = CAN_CONTINUE_SCENARIO
+
+    int_ctx.set_can_continue(ctx, pipeline, can_continue_flag)
+    int_ctx.set_confidence(ctx, pipeline, confidence)
+
+
+def facts(ctx, pipeline):
+    return provide_facts_request(ctx, pipeline)