edsl 0.1.33__py3-none-any.whl → 0.1.33.dev1__py3-none-any.whl

edsl/auto/StagePersonaDimensionValues.py

@@ -1,74 +0,0 @@
-from textwrap import dedent
-from dataclasses import dataclass
-
-from typing import List, Dict
-
-from edsl.auto.StageBase import StageBase
-from edsl.auto.StageBase import FlowDataBase
-
-from edsl.auto.StagePersonaDimensions import StagePersonaDimensions
-from edsl import Model
-from edsl.questions import QuestionList, QuestionExtract
-from edsl.scenarios import Scenario
-
-from edsl.auto.utilities import gen_pipeline
-
-
-class StagePersonaDimensionValues(StageBase):
-    input = StagePersonaDimensions.output
-
-    @dataclass
-    class Output(FlowDataBase):
-        attribute_results: List[str]
-        dimension_values: Dict[str, str]
-        persona: str
-
-    output = Output
-
-    def handle_data(self, data):
-        attribute_results = data.attribute_results
-        persona = data.persona
-        m = Model()
-        q = QuestionExtract(
-            question_text=dedent(
-                """\
-            This is a persona: "{{ persona }}"
-            They vary on the following dimensions: "{{ attribute_results }}"
-            For each dimenion, what are some values that this persona might have for that dimension?
-            Please keep answers very short, ideally one word.
-            """
-            ),
-            answer_template={k: None for k in attribute_results},
-            question_name="dimension_values",
-        )
-        results = (
-            q.by(Scenario({"attribute_results": attribute_results, "persona": persona}))
-            .by(m)
-            .run()
-        )
-        results.select("attribute_results", "dimension_values").print()
-        return self.output(
-            dimension_values=results.select("dimension_values").first(),
-            attribute_results=attribute_results,
-            persona=persona,
-        )
-
-
-if __name__ == "__main__":
-    from edsl.auto.StageQuestions import StageQuestions
-    from edsl.auto.StagePersona import StagePersona
-    from edsl.auto.StagePersonaDimensions import StagePersonaDimensions
-
-    pipeline = gen_pipeline(
-        [
-            StageQuestions,
-            StagePersona,
-            StagePersonaDimensions,
-            StagePersonaDimensionValues,
-        ]
-    )
-    pipeline.process(
-        pipeline.input(
-            overall_question="What are some factors that could determine whether someone likes ice cream?"
-        )
-    )

edsl/auto/StagePersonaDimensions.py

@@ -1,69 +0,0 @@
-from textwrap import dedent
-from dataclasses import dataclass
-
-from typing import List
-
-from edsl.auto.StageBase import StageBase
-from edsl.auto.StageBase import FlowDataBase
-
-from edsl.auto.StagePersona import StagePersona
-
-from edsl.questions import QuestionList
-from edsl.scenarios import Scenario
-from edsl import Model
-
-from edsl.auto.utilities import gen_pipeline
-
-
-class StagePersonaDimensions(StageBase):
-    input = StagePersona.output
-
-    @dataclass
-    class Output(FlowDataBase):
-        attribute_results: List[str]
-        persona: str
-
-    output = Output
-
-    def handle_data(self, data):
-        q_attributes = QuestionList(
-            question_text=dedent(
-                """\
-            Here is a persona: "{{ persona }}"
-            It was construced to be someone who could answer these questions: "{{ questions }}"
-
-            We want to identify the general dimensions that make up this persona.
-            E.g., if the person is desribed as 'happy' then a dimenion would be  'mood'
-            """
-            ),
-            question_name="find_attributes",
-        )
-        m = Model()
-        results = (
-            q_attributes.by(
-                Scenario({"persona": data.persona, "questions": data.questions})
-            )
-            .by(m)
-            .run()
-        )
-        (
-            results.select("find_attributes").print(
-                pretty_labels={
-                    "answer.find_attributes": f'Persona dimensions for: "{data.persona}"'
-                },
-                split_at_dot=False,
-            )
-        )
-        attribute_results = results.select("find_attributes").first()
-        return self.output(attribute_results=attribute_results, persona=data.persona)
-
-
-if __name__ == "__main__":
-    from edsl.auto.StageQuestions import StageQuestions
-
-    pipeline = gen_pipeline([StageQuestions, StagePersona, StagePersonaDimensions])
-    pipeline.process(
-        pipeline.input(
-            overall_question="What are some factors that could determine whether someone likes ice cream?"
-        )
-    )

edsl/auto/StageQuestions.py

@@ -1,73 +0,0 @@
-from dataclasses import dataclass
-from typing import List
-from textwrap import dedent
-
-
-from edsl import Scenario
-from edsl import Model
-from edsl.questions.QuestionList import QuestionList
-
-from edsl.auto.StageBase import StageBase
-from edsl.auto.StageBase import FlowDataBase
-
-from edsl.auto.utilities import gen_pipeline
-
-
-class StageQuestions(StageBase):
-    "This stages takes as input an overall question and returns a list of questions"
-
-    @dataclass
-    class Input(FlowDataBase):
-        overall_question: str
-        population: str
-
-    @dataclass
-    class Output(FlowDataBase):
-        questions: List[str]
-        population: str
-
-    input = Input
-    output = Output
-
-    def handle_data(self, data):
-        m = Model()
-        overall_question = data.overall_question
-        population = data.population
-        s = Scenario({"overall_question": overall_question, "population": population})
-        q = QuestionList(
-            question_text=dedent(
-                """\
-            Suppose I am interested in the question: 
-            "{{ overall_question }}" 
-            What would be some survey questions I could ask to {{ population }} that might shed light on this question?
-            """
-            ),
-            question_name="questions",
-        )
-        results = q.by(s).by(m).run()
-        (
-            results.select("questions").print(
-                pretty_labels={
-                    "answer.questions": f'Questions for overall question: "{overall_question }"'
-                },
-                split_at_dot=False,
-            )
-        )
-
-        raw_questions = results.select("questions").first()
-        questions = [q.replace("'", "").replace(":", "") for q in raw_questions]
-        return self.Output(questions=questions, population=population)
-
-
-if __name__ == "__main__":
-    pipeline = gen_pipeline([StageQuestions])
-
-    pipeline.process(
-        pipeline.input(
-            overall_question="What are some factors that could determine whether someone likes ice cream?",
-            population="Consumers",
-        )
-    )
-    StageQuestions.func(
-        overall_question="Why aren't my students studying more?", population="Tech"
-    )

edsl/auto/SurveyCreatorPipeline.py

@@ -1,21 +0,0 @@
-import random
-from typing import Dict, List, Any, TypeVar, Generator, Optional
-
-from textwrap import dedent
-
-# from edsl.language_models.model_interfaces.LanguageModelOpenAIFour import LanguageModelOpenAIFour
-from edsl import Model
-from edsl.agents.AgentList import AgentList
-from edsl.results.Results import Results
-from edsl import Agent
-
-from edsl import Scenario
-from edsl.surveys.Survey import Survey
-
-from edsl.questions.QuestionMultipleChoice import QuestionMultipleChoice
-from edsl.questions.QuestionFreeText import QuestionFreeText
-from edsl.auto.utilities import gen_pipeline
-from edsl.conjure.naming_utilities import sanitize_string
-
-
-m = Model()

edsl/auto/utilities.py

@@ -1,224 +0,0 @@
-from textwrap import dedent
-import random
-from typing import List, TypeVar, Generator, Optional
-from edsl.auto.StageBase import StageBase
-from edsl.conjure.naming_utilities import sanitize_string
-from edsl import Agent, Survey, Model, Cache, AgentList
-from edsl import QuestionFreeText, Scenario
-from edsl import QuestionMultipleChoice, Scenario, Agent, ScenarioList
-
-StageClassType = TypeVar("StageClassType", bound=StageBase)
-
-
-def gen_pipeline(stages_list: List[StageClassType]) -> StageBase:
-    """Takes as input a list of Stage classes & returns a pipeline of instantiated stages.
-    A pipeline is a linked list of stages where each stage has a next_stage attribute.
-
-    """
-    pipeline = stages_list[0]()
-    last_stage = pipeline
-    for stage in stages_list[1:]:
-        while last_stage.next_stage is not None:  # find the end of the pipeline
-            last_stage = last_stage.next_stage
-        stage_to_add = stage()
-        last_stage.next_stage = stage_to_add
-    return pipeline
-
-
-q_eligibility = QuestionMultipleChoice(
-    question_text=dedent(
-        """\
-        Consider this set of question: '{{ questions }}'.
-        Consider this persona: '{{ persona }}'.
-        Would this persona be able to answer all of these questions?
-        """
-    ),
-    question_options=["No", "Yes"],
-    question_name="eligibility",
-)
-
-
-def agent_list_eligibility(
-    agent_list: AgentList,
-    survey: Optional[Survey] = None,
-    model: Optional[Model] = None,
-    cache: Optional[Cache] = None,
-) -> List[bool]:
-    """
-    Returns whether each agent in a list is elgible for a survey i.e., can answer every question.
-
-    >>> from edsl.language_models import LanguageModel
-    >>> m = LanguageModel.example(canned_response = "1", test_model = True)
-    >>> agent_list_eligibility(AgentList.example())
-    [True, True]
-    >>> agent_list_eligibility(AgentList.example().add_trait('persona', 2*["Cool dude"]), survey = Survey.example(), model = m)
-    [True, True]
-    """
-    if survey is None:
-        return [True] * len(agent_list)
-    if "persona" not in agent_list.all_traits:
-        raise ValueError(
-            f"Each agent needs to have a persona attribute; traits are {agent_list.all_traits}"
-        )
-    sl = agent_list.select("persona").to_scenario_list()
-    sl.add_value("questions", [q.question_text for q in survey._questions])
-    results = q_eligibility.by(sl).by(model).run(cache=cache)
-    return [r == "Yes" for r in results.select("eligibility").to_list()]
-
-
-def agent_eligibility(
-    agent: Agent,
-    survey: Survey,
-    model: Optional[Model] = None,
-    cache: Optional[Cache] = None,
-) -> bool:
-    """NB: This could be parallelized.
-
-    >>> from edsl.language_models import LanguageModel
-    >>> m = LanguageModel.example(canned_response = "1", test_model = True)
-    >>> agent_eligibility(agent = Agent.example().add_trait({'persona': "Persona"}), survey = Survey.example(), model = m)
-    True
-
-    """
-    model = model or Model()
-
-    questions = [q.question_text for q in survey._questions]
-    persona = agent.traits["persona"]
-    return (
-        q_eligibility(model=model, questions=questions, persona=persona, cache=cache)
-        == "Yes"
-    )
-    # results = (
-    #     q.by(model)
-    #     .by(Scenario({"questions": questions, "persona": persona}))
-    #     .run(cache=cache)
-    # )
-    # return results.select("eligibility").first() == "Yes"
-
-
-def gen_agent_traits(dimension_dict: dict, seed_value: Optional[str] = None):
-    """
-    >>> dimension_dict = {'attitude':['positive', 'negative']}
-    >>> ag = gen_agent_traits(dimension_dict)
-    >>> a = next(ag)
-    >>> a == {'attitude': 'positive'} or a == {'attitude': 'negative'}
-    True
-    >>> len([next(ag) for _ in range(100)])
-    100
-    """
-    if seed_value is None:
-        seed_value = "edsl"
-
-    random.seed(seed_value)
-
-    while True:
-        new_agent_traits = {}
-        for key, list_of_values in dimension_dict.items():
-            new_agent_traits[key] = random.choice(list_of_values)
-        yield new_agent_traits
-
-
-def agent_generator(
-    persona: str,
-    dimension_dict: dict,
-    model: Optional[Model] = None,
-    cache: Optional["Cache"] = None,
-) -> Generator["Results", None, None]:
-    """
-    >>> from edsl.language_models import LanguageModel
-    >>> m = LanguageModel.example(canned_response = "This is a cool dude.", test_model = True)
-    >>> ag = agent_generator(persona = "Base person", dimension_dict = {'attitude':['Positive', 'Negative']}, model = m)
-    >>> next(ag).select('new_agent_persona').first()
-    'This is a cool dude.'
-    >>> next(ag).select('new_agent_persona').first()
-    'This is a cool dude.'
-    """
-
-    if model is None:
-        model = Model()
-
-    q = QuestionFreeText(
-        question_text=dedent(
-            """\
-    Consider this persona: '{{ persona }}'.
-    Now imagine writing a new persona with these traits: 
-    '{{ new_agent_traits }}'
-    Please write this persona as a narrative.
-    """
-        ),
-        question_name="new_agent_persona",
-    )
-    agent_trait_generator = gen_agent_traits(dimension_dict)
-    codebook = {sanitize_string(k): k for k in dimension_dict.keys()}
-    while True:
-        new_agent_traits = next(agent_trait_generator)
-        yield q(
-            persona=persona,
-            new_agent_traits=new_agent_traits,
-            codebook=codebook,
-            just_answer=False,
-            cache=cache,
-            model=model,
-        )
-
-
-def create_agents(
-    agent_generator: Generator["Results", None, None],
-    survey: Optional[Survey] = None,
-    num_agents=11,
-) -> AgentList:
-    """
-    >>> from edsl.language_models import LanguageModel
-    >>> m = LanguageModel.example(canned_response = "This is a cool dude.", test_model = True)
-    >>> ag = agent_generator(persona = "Base person", dimension_dict = {'attitude':['Positive', 'Negative']}, model = m)
-    >>> new_agent_list = create_agents(agent_generator = ag)
-    >>> new_agent_list
-
-    """
-    agent_list = AgentList([])
-
-    MAX_ITERATIONS_MULTIPLIER = 2
-    iterations = 0
-
-    while len(agent_list) < num_agents:
-        iterations += 1
-        candidate_agent = next(agent_generator)
-        codebook = candidate_agent.select("codebook").to_list()[0]
-
-        koobedoc = {v: k for k, v in codebook.items()}
-        persona = candidate_agent.select("new_agent_persona").to_list()[0]
-        traits = candidate_agent.select("new_agent_traits").to_list()[0]
-        new_traits = {koobedoc[key]: value for key, value in traits.items()} | {
-            "persona": persona
-        }
-        agent = Agent(traits=new_traits, codebook=codebook)
-        if survey is not None:
-            if agent_eligibility(agent, survey):
-                agent_list.append(agent)
-            else:
-                print("Agent not eligible")
-        else:
-            agent_list.append(agent)
-
-        if iterations > MAX_ITERATIONS_MULTIPLIER * num_agents:
-            raise Exception("Too many failures")
-
-    return agent_list
-
-
-if __name__ == "__main__":
-    import doctest
-
-    doctest.testmod()
-    # from edsl.language_models import LanguageModel
-
-    # m = LanguageModel.example(canned_response="This is a cool dude.", test_model=True)
-    # ag = agent_generator(
-    #     persona="Base person",
-    #     dimension_dict={"attitude": ["Positive", "Negative"]},
-    #     model=m,
-    # )
-    # example = [next(ag).select("new_agent_persona").first() for _ in range(10)]
-    # dimension_dict = {"attitude": ["positive", "negative"]}
-    # ag = gen_agent_traits(dimension_dict)
-    # example = [next(ag) for _ in range(100)]

edsl/coop/PriceFetcher.py

@@ -1,58 +0,0 @@
-import requests
-import csv
-from io import StringIO
-
-
-class PriceFetcher:
-    _instance = None
-
-    def __new__(cls):
-        if cls._instance is None:
-            cls._instance = super(PriceFetcher, cls).__new__(cls)
-            cls._instance._cached_prices = None
-        return cls._instance
-
-    def fetch_prices(self):
-        if self._cached_prices is not None:
-            return self._cached_prices
-
-        import requests
-        import csv
-        from io import StringIO
-
-        sheet_id = "1SAO3Bhntefl0XQHJv27rMxpvu6uzKDWNXFHRa7jrUDs"
-
-        # Construct the URL to fetch the CSV
-        url = f"https://docs.google.com/spreadsheets/d/{sheet_id}/export?format=csv"
-
-        try:
-            # Fetch the CSV data
-            response = requests.get(url)
-            response.raise_for_status()  # Raise an exception for bad responses
-
-            # Parse the CSV data
-            csv_data = StringIO(response.text)
-            reader = csv.reader(csv_data)
-
-            # Convert to list of dictionaries
-            headers = next(reader)
-            data = [dict(zip(headers, row)) for row in reader]
-
-            # self._cached_prices = data
-            # return data
-            price_lookup = {}
-            for entry in data:
-                service = entry.get("service", None)
-                model = entry.get("model", None)
-                if service and model:
-                    token_type = entry.get("token_type", None)
-                    if (service, model) in price_lookup:
-                        price_lookup[(service, model)].update({token_type: entry})
-                    else:
-                        price_lookup[(service, model)] = {token_type: entry}
-            self._cached_prices = price_lookup
-            return self._cached_prices
-
-        except requests.RequestException as e:
-            # print(f"An error occurred: {e}")
-            return {}

edsl/inference_services/MistralAIService.py

@@ -1,120 +0,0 @@
-import os
-from typing import Any, List
-from edsl.inference_services.InferenceServiceABC import InferenceServiceABC
-from edsl.language_models.LanguageModel import LanguageModel
-import asyncio
-from mistralai import Mistral
-
-from edsl.exceptions.language_models import LanguageModelBadResponseError
-
-
-class MistralAIService(InferenceServiceABC):
-    """Mistral AI service class."""
-
-    key_sequence = ["choices", 0, "message", "content"]
-    usage_sequence = ["usage"]
-
-    _inference_service_ = "mistral"
-    _env_key_name_ = "MISTRAL_API_KEY"  # Environment variable for Mistral API key
-    input_token_name = "prompt_tokens"
-    output_token_name = "completion_tokens"
-
-    _sync_client_instance = None
-    _async_client_instance = None
-
-    _sync_client = Mistral
-    _async_client = Mistral
-
-    _models_list_cache: List[str] = []
-    model_exclude_list = []
-
-    def __init_subclass__(cls, **kwargs):
-        super().__init_subclass__(**kwargs)
-        # so subclasses have to create their own instances of the clients
-        cls._sync_client_instance = None
-        cls._async_client_instance = None
-
-    @classmethod
-    def sync_client(cls):
-        if cls._sync_client_instance is None:
-            cls._sync_client_instance = cls._sync_client(
-                api_key=os.getenv(cls._env_key_name_)
-            )
-        return cls._sync_client_instance
-
-    @classmethod
-    def async_client(cls):
-        if cls._async_client_instance is None:
-            cls._async_client_instance = cls._async_client(
-                api_key=os.getenv(cls._env_key_name_)
-            )
-        return cls._async_client_instance
-
-    @classmethod
-    def available(cls) -> list[str]:
-        if not cls._models_list_cache:
-            cls._models_list_cache = [
-                m.id for m in cls.sync_client().models.list().data
-            ]
-
-        return cls._models_list_cache
-
-    @classmethod
-    def create_model(
-        cls, model_name: str = "mistral", model_class_name=None
-    ) -> LanguageModel:
-        if model_class_name is None:
-            model_class_name = cls.to_class_name(model_name)
-
-        class LLM(LanguageModel):
-            """
-            Child class of LanguageModel for interacting with Mistral models.
-            """
-
-            key_sequence = cls.key_sequence
-            usage_sequence = cls.usage_sequence
-
-            input_token_name = cls.input_token_name
-            output_token_name = cls.output_token_name
-
-            _inference_service_ = cls._inference_service_
-            _model_ = model_name
-            _parameters_ = {
-                "temperature": 0.5,
-                "max_tokens": 512,
-                "top_p": 0.9,
-            }
-
-            _tpm = cls.get_tpm(cls)
-            _rpm = cls.get_rpm(cls)
-
-            def sync_client(self):
-                return cls.sync_client()
-
-            def async_client(self):
-                return cls.async_client()
-
-            async def async_execute_model_call(
-                self, user_prompt: str, system_prompt: str = ""
-            ) -> dict[str, Any]:
-                """Calls the Mistral API and returns the API response."""
-                s = self.async_client()
-
-                try:
-                    res = await s.chat.complete_async(
-                        model=model_name,
-                        messages=[
-                            {
-                                "content": user_prompt,
-                                "role": "user",
-                            },
-                        ],
-                    )
-                except Exception as e:
-                    raise LanguageModelBadResponseError(f"Error with Mistral API: {e}")
-
-                return res.model_dump()
-
-        LLM.__name__ = model_class_name
-
-        return LLM