Fastapi refactor update
This commit is contained in:
Carlos-Mesquita
417
app/services/impl/exam/level/level.py
Normal file
417
app/services/impl/exam/level/level.py
Normal file
@@ -0,0 +1,417 @@
|
||||
import json
|
||||
import random
|
||||
import uuid
|
||||
|
||||
from typing import Dict
|
||||
|
||||
from fastapi import UploadFile
|
||||
|
||||
from app.configs.constants import GPTModels, TemperatureSettings, EducationalContent
|
||||
from app.helpers import ExercisesHelper
|
||||
from app.repositories.abc import IDocumentStore
|
||||
from app.services.abc import ILevelService, ILLMService, IReadingService, IWritingService, ISpeakingService, \
|
||||
IListeningService
|
||||
from .custom import CustomLevelModule
|
||||
from .upload import UploadLevelModule
|
||||
|
||||
|
||||
class LevelService(ILevelService):
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
llm: ILLMService,
|
||||
document_store: IDocumentStore,
|
||||
mc_variants: Dict,
|
||||
reading_service: IReadingService,
|
||||
writing_service: IWritingService,
|
||||
speaking_service: ISpeakingService,
|
||||
listening_service: IListeningService
|
||||
):
|
||||
self._llm = llm
|
||||
self._document_store = document_store
|
||||
self._reading_service = reading_service
|
||||
self._custom_module = CustomLevelModule(
|
||||
llm, self, reading_service, listening_service, writing_service, speaking_service
|
||||
)
|
||||
self._upload_module = UploadLevelModule(llm)
|
||||
|
||||
# TODO: normal and blank spaces only differ on "multiple choice blank space questions" in the prompt
|
||||
# mc_variants are stored in ./mc_variants.json
|
||||
self._mc_variants = mc_variants
|
||||
|
||||
async def upload_level(self, upload: UploadFile) -> Dict:
|
||||
return await self._upload_module.generate_level_from_file(upload)
|
||||
|
||||
async def get_custom_level(self, data: Dict):
|
||||
return await self._custom_module.get_custom_level(data)
|
||||
|
||||
async def get_level_exam(
|
||||
self, number_of_exercises: int = 25, min_timer: int = 25, diagnostic: bool = False
|
||||
) -> Dict:
|
||||
exercises = await self.gen_multiple_choice("normal", number_of_exercises, utas=False)
|
||||
return {
|
||||
"exercises": [exercises],
|
||||
"isDiagnostic": diagnostic,
|
||||
"minTimer": min_timer,
|
||||
"module": "level"
|
||||
}
|
||||
|
||||
async def get_level_utas(self, diagnostic: bool = False, min_timer: int = 25):
|
||||
# Formats
|
||||
mc = {
|
||||
"id": str(uuid.uuid4()),
|
||||
"prompt": "Choose the correct word or group of words that completes the sentences.",
|
||||
"questions": None,
|
||||
"type": "multipleChoice",
|
||||
"part": 1
|
||||
}
|
||||
|
||||
umc = {
|
||||
"id": str(uuid.uuid4()),
|
||||
"prompt": "Choose the underlined word or group of words that is not correct.",
|
||||
"questions": None,
|
||||
"type": "multipleChoice",
|
||||
"part": 2
|
||||
}
|
||||
|
||||
bs_1 = {
|
||||
"id": str(uuid.uuid4()),
|
||||
"prompt": "Read the text and write the correct word for each space.",
|
||||
"questions": None,
|
||||
"type": "blankSpaceText",
|
||||
"part": 3
|
||||
}
|
||||
|
||||
bs_2 = {
|
||||
"id": str(uuid.uuid4()),
|
||||
"prompt": "Read the text and write the correct word for each space.",
|
||||
"questions": None,
|
||||
"type": "blankSpaceText",
|
||||
"part": 4
|
||||
}
|
||||
|
||||
reading = {
|
||||
"id": str(uuid.uuid4()),
|
||||
"prompt": "Read the text and answer the questions below.",
|
||||
"questions": None,
|
||||
"type": "readingExercises",
|
||||
"part": 5
|
||||
}
|
||||
|
||||
all_mc_questions = []
|
||||
|
||||
# PART 1
|
||||
# await self._gen_multiple_choice("normal", number_of_exercises, utas=False)
|
||||
mc_exercises1 = await self.gen_multiple_choice(
|
||||
"blank_space", 15, 1, utas=True, all_exams=all_mc_questions
|
||||
)
|
||||
print(json.dumps(mc_exercises1, indent=4))
|
||||
all_mc_questions.append(mc_exercises1)
|
||||
|
||||
# PART 2
|
||||
mc_exercises2 = await self.gen_multiple_choice(
|
||||
"blank_space", 15, 16, utas=True, all_exams=all_mc_questions
|
||||
)
|
||||
print(json.dumps(mc_exercises2, indent=4))
|
||||
all_mc_questions.append(mc_exercises2)
|
||||
|
||||
# PART 3
|
||||
mc_exercises3 = await self.gen_multiple_choice(
|
||||
"blank_space", 15, 31, utas=True, all_exams=all_mc_questions
|
||||
)
|
||||
print(json.dumps(mc_exercises3, indent=4))
|
||||
all_mc_questions.append(mc_exercises3)
|
||||
|
||||
mc_exercises = mc_exercises1['questions'] + mc_exercises2['questions'] + mc_exercises3['questions']
|
||||
print(json.dumps(mc_exercises, indent=4))
|
||||
mc["questions"] = mc_exercises
|
||||
|
||||
# Underlined mc
|
||||
underlined_mc = await self.gen_multiple_choice(
|
||||
"underline", 15, 46, utas=True, all_exams=all_mc_questions
|
||||
)
|
||||
print(json.dumps(underlined_mc, indent=4))
|
||||
umc["questions"] = underlined_mc
|
||||
|
||||
# Blank Space text 1
|
||||
blank_space_text_1 = await self.gen_blank_space_text_utas(12, 61, 250)
|
||||
print(json.dumps(blank_space_text_1, indent=4))
|
||||
bs_1["questions"] = blank_space_text_1
|
||||
|
||||
# Blank Space text 2
|
||||
blank_space_text_2 = await self.gen_blank_space_text_utas(14, 73, 350)
|
||||
print(json.dumps(blank_space_text_2, indent=4))
|
||||
bs_2["questions"] = blank_space_text_2
|
||||
|
||||
# Reading text
|
||||
reading_text = await self.gen_reading_passage_utas(87, 10, 4)
|
||||
print(json.dumps(reading_text, indent=4))
|
||||
reading["questions"] = reading_text
|
||||
|
||||
return {
|
||||
"exercises": {
|
||||
"blankSpaceMultipleChoice": mc,
|
||||
"underlinedMultipleChoice": umc,
|
||||
"blankSpaceText1": bs_1,
|
||||
"blankSpaceText2": bs_2,
|
||||
"readingExercises": reading,
|
||||
},
|
||||
"isDiagnostic": diagnostic,
|
||||
"minTimer": min_timer,
|
||||
"module": "level"
|
||||
}
|
||||
|
||||
async def gen_multiple_choice(
|
||||
self, mc_variant: str, quantity: int, start_id: int = 1, *, utas: bool = False, all_exams=None
|
||||
):
|
||||
mc_template = self._mc_variants[mc_variant]
|
||||
blank_mod = " blank space " if mc_variant == "blank_space" else " "
|
||||
|
||||
gen_multiple_choice_for_text: str = (
|
||||
'Generate {quantity} multiple choice{blank}questions of 4 options for an english level exam, some easy '
|
||||
'questions, some intermediate questions and some advanced questions. Ensure that the questions cover '
|
||||
'a range of topics such as verb tense, subject-verb agreement, pronoun usage, sentence structure, and '
|
||||
'punctuation. Make sure every question only has 1 correct answer.'
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "system",
|
||||
"content": (
|
||||
f'You are a helpful assistant designed to output JSON on this format: {mc_template}'
|
||||
)
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": gen_multiple_choice_for_text.format(quantity=str(quantity), blank=blank_mod)
|
||||
}
|
||||
]
|
||||
|
||||
if mc_variant == "underline":
|
||||
messages.append({
|
||||
"role": "user",
|
||||
"content": (
|
||||
'The type of multiple choice in the prompt has wrong words or group of words and the options '
|
||||
'are to find the wrong word or group of words that are underlined in the prompt. \nExample:\n'
|
||||
'Prompt: "I <u>complain</u> about my boss <u>all the time</u>, but my colleagues <u>thinks</u> '
|
||||
'the boss <u>is</u> nice."\n'
|
||||
'Options:\na: "complain"\nb: "all the time"\nc: "thinks"\nd: "is"'
|
||||
)
|
||||
})
|
||||
|
||||
question = await self._llm.prediction(
|
||||
GPTModels.GPT_4_O, messages, ["questions"], TemperatureSettings.GEN_QUESTION_TEMPERATURE
|
||||
)
|
||||
|
||||
if len(question["questions"]) != quantity:
|
||||
return await self.gen_multiple_choice(mc_variant, quantity, start_id, utas=utas, all_exams=all_exams)
|
||||
else:
|
||||
if not utas:
|
||||
all_exams = await self._document_store.get_all("level")
|
||||
seen_keys = set()
|
||||
for i in range(len(question["questions"])):
|
||||
question["questions"][i], seen_keys = await self._replace_exercise_if_exists(
|
||||
all_exams, question["questions"][i], question, seen_keys, mc_variant, utas
|
||||
)
|
||||
return {
|
||||
"id": str(uuid.uuid4()),
|
||||
"prompt": "Select the appropriate option.",
|
||||
"questions": ExercisesHelper.fix_exercise_ids(question, start_id)["questions"],
|
||||
"type": "multipleChoice",
|
||||
}
|
||||
else:
|
||||
if all_exams is not None:
|
||||
seen_keys = set()
|
||||
for i in range(len(question["questions"])):
|
||||
question["questions"][i], seen_keys = await self._replace_exercise_if_exists(
|
||||
all_exams, question["questions"][i], question, seen_keys, mc_variant, utas
|
||||
)
|
||||
response = ExercisesHelper.fix_exercise_ids(question, start_id)
|
||||
response["questions"] = ExercisesHelper.randomize_mc_options_order(response["questions"])
|
||||
return response
|
||||
|
||||
async def _generate_single_multiple_choice(self, mc_variant: str = "normal"):
|
||||
mc_template = self._mc_variants[mc_variant]["questions"][0]
|
||||
blank_mod = " blank space " if mc_variant == "blank_space" else " "
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "system",
|
||||
"content": (
|
||||
f'You are a helpful assistant designed to output JSON on this format: {mc_template}'
|
||||
)
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": (
|
||||
f'Generate 1 multiple choice {blank_mod} question of 4 options for an english level exam, '
|
||||
f'it can be easy, intermediate or advanced.'
|
||||
)
|
||||
|
||||
}
|
||||
]
|
||||
|
||||
if mc_variant == "underline":
|
||||
messages.append({
|
||||
"role": "user",
|
||||
"content": (
|
||||
'The type of multiple choice in the prompt has wrong words or group of words and the options '
|
||||
'are to find the wrong word or group of words that are underlined in the prompt. \nExample:\n'
|
||||
'Prompt: "I <u>complain</u> about my boss <u>all the time</u>, but my colleagues <u>thinks</u> '
|
||||
'the boss <u>is</u> nice."\n'
|
||||
'Options:\na: "complain"\nb: "all the time"\nc: "thinks"\nd: "is"'
|
||||
)
|
||||
})
|
||||
|
||||
question = await self._llm.prediction(
|
||||
GPTModels.GPT_4_O, messages, ["options"], TemperatureSettings.GEN_QUESTION_TEMPERATURE
|
||||
)
|
||||
|
||||
return question
|
||||
|
||||
async def _replace_exercise_if_exists(
|
||||
self, all_exams, current_exercise, current_exam, seen_keys, mc_variant: str, utas: bool = False
|
||||
):
|
||||
# Extracting relevant fields for comparison
|
||||
key = (current_exercise['prompt'], tuple(sorted(option['text'] for option in current_exercise['options'])))
|
||||
# Check if the key is in the set
|
||||
if key in seen_keys:
|
||||
return await self._replace_exercise_if_exists(
|
||||
all_exams, await self._generate_single_multiple_choice(mc_variant), current_exam, seen_keys,
|
||||
mc_variant, utas
|
||||
)
|
||||
else:
|
||||
seen_keys.add(key)
|
||||
|
||||
if not utas:
|
||||
for exam in all_exams:
|
||||
exam_dict = exam.to_dict()
|
||||
if len(exam_dict.get("parts", [])) > 0:
|
||||
exercise_dict = exam_dict.get("parts", [])[0]
|
||||
if len(exercise_dict.get("exercises", [])) > 0:
|
||||
if any(
|
||||
exercise["prompt"] == current_exercise["prompt"] and
|
||||
any(exercise["options"][0]["text"] == current_option["text"] for current_option in
|
||||
current_exercise["options"])
|
||||
for exercise in exercise_dict.get("exercises", [])[0]["questions"]
|
||||
):
|
||||
return await self._replace_exercise_if_exists(
|
||||
all_exams, await self._generate_single_multiple_choice(mc_variant), current_exam,
|
||||
seen_keys, mc_variant, utas
|
||||
)
|
||||
else:
|
||||
for exam in all_exams:
|
||||
if any(
|
||||
exercise["prompt"] == current_exercise["prompt"] and
|
||||
any(exercise["options"][0]["text"] == current_option["text"] for current_option in
|
||||
current_exercise["options"])
|
||||
for exercise in exam.get("questions", [])
|
||||
):
|
||||
return await self._replace_exercise_if_exists(
|
||||
all_exams, await self._generate_single_multiple_choice(mc_variant), current_exam,
|
||||
seen_keys, mc_variant, utas
|
||||
)
|
||||
return current_exercise, seen_keys
|
||||
|
||||
async def gen_blank_space_text_utas(
|
||||
self, quantity: int, start_id: int, size: int, topic=random.choice(EducationalContent.MTI_TOPICS)
|
||||
):
|
||||
json_template = self._mc_variants["blank_space_text"]
|
||||
messages = [
|
||||
{
|
||||
"role": "system",
|
||||
"content": f'You are a helpful assistant designed to output JSON on this format: {json_template}'
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": f'Generate a text of at least {size} words about the topic {topic}.'
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": (
|
||||
f'From the generated text choose {quantity} words (cannot be sequential words) to replace '
|
||||
'once with {{id}} where id starts on ' + str(start_id) + ' and is incremented for each word. '
|
||||
'The ids must be ordered throughout the text and the words must be replaced only once. '
|
||||
'Put the removed words and respective ids on the words array of the json in the correct order.'
|
||||
)
|
||||
}
|
||||
]
|
||||
|
||||
question = await self._llm.prediction(
|
||||
GPTModels.GPT_4_O, messages, ["question"], TemperatureSettings.GEN_QUESTION_TEMPERATURE
|
||||
)
|
||||
|
||||
return question["question"]
|
||||
|
||||
async def gen_reading_passage_utas(
|
||||
self, start_id, sa_quantity: int, mc_quantity: int, topic=random.choice(EducationalContent.MTI_TOPICS)
|
||||
):
|
||||
passage = await self._reading_service.generate_reading_passage(1, topic)
|
||||
short_answer = await self._gen_short_answer_utas(passage["text"], start_id, sa_quantity)
|
||||
mc_exercises = await self._gen_text_multiple_choice_utas(passage["text"], start_id + sa_quantity, mc_quantity)
|
||||
return {
|
||||
"exercises": {
|
||||
"shortAnswer": short_answer,
|
||||
"multipleChoice": mc_exercises,
|
||||
},
|
||||
"text": {
|
||||
"content": passage["text"],
|
||||
"title": passage["title"]
|
||||
}
|
||||
}
|
||||
|
||||
async def _gen_short_answer_utas(self, text: str, start_id: int, sa_quantity: int):
|
||||
json_format = {"questions": [{"id": 1, "question": "question", "possible_answers": ["answer_1", "answer_2"]}]}
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "system",
|
||||
"content": f'You are a helpful assistant designed to output JSON on this format: {json_format}'
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": (
|
||||
f'Generate {sa_quantity} short answer questions, and the possible answers, must have '
|
||||
f'maximum 3 words per answer, about this text:\n"{text}"'
|
||||
)
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": f'The id starts at {start_id}.'
|
||||
}
|
||||
]
|
||||
|
||||
question = await self._llm.prediction(
|
||||
GPTModels.GPT_4_O, messages, ["questions"], TemperatureSettings.GEN_QUESTION_TEMPERATURE
|
||||
)
|
||||
|
||||
return question["questions"]
|
||||
|
||||
async def _gen_text_multiple_choice_utas(self, text: str, start_id: int, mc_quantity: int):
|
||||
json_template = self._mc_variants["text_mc_utas"]
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "system",
|
||||
"content": f'You are a helpful assistant designed to output JSON on this format: {json_template}'
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": f'Generate {mc_quantity} multiple choice questions of 4 options for this text:\n{text}'
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": 'Make sure every question only has 1 correct answer.'
|
||||
}
|
||||
]
|
||||
|
||||
question = await self._llm.prediction(
|
||||
GPTModels.GPT_4_O, messages, ["questions"], TemperatureSettings.GEN_QUESTION_TEMPERATURE
|
||||
)
|
||||
|
||||
if len(question["questions"]) != mc_quantity:
|
||||
return await self._gen_text_multiple_choice_utas(text, mc_quantity, start_id)
|
||||
else:
|
||||
response = ExercisesHelper.fix_exercise_ids(question, start_id)
|
||||
response["questions"] = ExercisesHelper.randomize_mc_options_order(response["questions"])
|
||||
return response
|
||||
Reference in New Issue
Block a user