edsl 0.1.38.dev4__py3-none-any.whl → 0.1.39__py3-none-any.whl

edsl/results/Result.py

@@ -1,81 +1,61 @@
 # """This module contains the Result class, which captures the result of one interview."""
 from __future__ import annotations
+import inspect
 from collections import UserDict
-from typing import Any, Type, Callable, Optional
-from collections import UserDict
+from typing import Any, Type, Callable, Optional, TYPE_CHECKING, Union
 from edsl.Base import Base
 from edsl.utilities.decorators import add_edsl_version, remove_edsl_version
 
+if TYPE_CHECKING:
+    from edsl.agents.Agent import Agent
+    from edsl.scenarios.Scenario import Scenario
+    from edsl.language_models.LanguageModel import LanguageModel
+    from edsl.prompts.Prompt import Prompt
+    from edsl.surveys.Survey import Survey
 
-class PromptDict(UserDict):
-    """A dictionary that is used to store the prompt for a given result."""
-
-    def rich_print(self):
-        """Display an object as a table."""
-        from rich.table import Table
-
-        table = Table(title="")
-        table.add_column("Attribute", style="bold")
-        table.add_column("Value")
 
-        to_display = self
-        for attr_name, attr_value in to_display.items():
-            table.add_row(attr_name, repr(attr_value))
+QuestionName = str
+AnswerValue = Any
 
-        return table
 
+class AgentNamer:
+    """Maintains a registry of agent names to ensure unique naming."""
 
-def agent_namer_closure():
-    """Return a function that can be used to name an agent."""
-    agent_dict = {}
+    def __init__(self):
+        self._registry = {}
 
-    def agent_namer(agent):
-        """Return a name for an agent. If the agent has been named before, return the same name. Otherwise, return a new name."""
-        nonlocal agent_dict
-        agent_count = len(agent_dict)
-        if id(agent) in agent_dict:
-            return agent_dict[id(agent)]
-        else:
-            agent_dict[id(agent)] = f"Agent_{agent_count}"
-            return agent_dict[id(agent)]
-
-    return agent_namer
+    def get_name(self, agent: "Agent") -> str:
+        """Get or create a unique name for an agent."""
+        agent_id = id(agent)
+        if agent_id not in self._registry:
+            self._registry[agent_id] = f"Agent_{len(self._registry)}"
+        return self._registry[agent_id]
 
 
-agent_namer = agent_namer_closure()
+# Global instance for agent naming
+agent_namer = AgentNamer().get_name
 
 
 class Result(Base, UserDict):
     """
     This class captures the result of one interview.
-
-    The answer dictionary has the structure:
-
-    >>> import warnings
-    >>> warnings.simplefilter("ignore", UserWarning)
-    >>> Result.example().answer == {'how_feeling_yesterday': 'Great', 'how_feeling': 'OK'}
-    True
-
-    Its main data is an Agent, a Scenario, a Model, an Iteration, and an Answer.
-    These are stored both in the UserDict and as attributes.
-
-
     """
 
     def __init__(
         self,
         agent: "Agent",
         scenario: "Scenario",
-        model: Type["LanguageModel"],
+        model: "LanguageModel",
         iteration: int,
-        answer: str,
-        prompt: dict[str, str] = None,
-        raw_model_response=None,
+        answer: dict[QuestionName, AnswerValue],
+        prompt: dict[QuestionName, str] = None,
+        raw_model_response: Optional[dict] = None,
         survey: Optional["Survey"] = None,
-        question_to_attributes: Optional[dict] = None,
+        question_to_attributes: Optional[dict[QuestionName, Any]] = None,
         generated_tokens: Optional[dict] = None,
         comments_dict: Optional[dict] = None,
-        cache_used_dict: Optional[dict] = None,
+        cache_used_dict: Optional[dict[QuestionName, bool]] = None,
+        indices: Optional[dict] = None,
     ):
         """Initialize a Result object.
 
@@ -86,26 +66,17 @@ class Result(Base, UserDict):
         :param answer: The answer string.
         :param prompt: A dictionary of prompts.
         :param raw_model_response: The raw model response.
+        :param survey: The Survey object.
+        :param question_to_attributes: A dictionary of question attributes.
+        :param generated_tokens: A dictionary of generated tokens.
+        :param comments_dict: A dictionary of comments.
+        :param cache_used_dict: A dictionary of cache usage.
+        :param indices: A dictionary of indices.
 
         """
-        if question_to_attributes is not None:
-            question_to_attributes = question_to_attributes
-        else:
-            question_to_attributes = {}
-
-        if survey is not None:
-            question_to_attributes = {
-                q.question_name: {
-                    "question_text": q.question_text,
-                    "question_type": q.question_type,
-                    "question_options": (
-                        None
-                        if not hasattr(q, "question_options")
-                        else q.question_options
-                    ),
-                }
-                for q in survey.questions
-            }
+        self.question_to_attributes = (
+            question_to_attributes or self._create_question_to_attributes(survey)
+        )
 
         data = {
             "agent": agent,
@@ -118,81 +89,127 @@ class Result(Base, UserDict):
             "question_to_attributes": question_to_attributes,
             "generated_tokens": generated_tokens or {},
             "comments_dict": comments_dict or {},
+            "cache_used_dict": cache_used_dict or {},
         }
         super().__init__(**data)
-        # but also store the data as attributes
-        self.agent = agent
-        self.scenario = scenario
-        self.model = model
-        self.iteration = iteration
-        self.answer = answer
-        self.prompt = prompt or {}
-        self.raw_model_response = raw_model_response or {}
-        self.survey = survey
-        self.question_to_attributes = question_to_attributes
-        self.generated_tokens = generated_tokens
-        self.comments_dict = comments_dict or {}
-        self.cache_used_dict = cache_used_dict or {}
-
-        self._combined_dict = None
-        self._problem_keys = None
-
-    def _repr_html_(self):
-        # d = self.to_dict(add_edsl_version=False)
-        d = self.to_dict(add_edsl_version=False)
-        data = [[k, v] for k, v in d.items()]
-        from tabulate import tabulate
-
-        table = str(tabulate(data, headers=["keys", "values"], tablefmt="html"))
-        return f"<pre>{table}</pre>"
-
-    ###############
-    # Used in Results
-    ###############
+        self.indices = indices
+        self._sub_dicts = self._construct_sub_dicts()
+        (
+            self._combined_dict,
+            self._problem_keys,
+        ) = self._compute_combined_dict_and_problem_keys()
+
+    @staticmethod
+    def _create_question_to_attributes(survey):
+        """Create a dictionary of question attributes."""
+        if survey is None:
+            return {}
+        return {
+            q.question_name: {
+                "question_text": q.question_text,
+                "question_type": q.question_type,
+                "question_options": (
+                    None if not hasattr(q, "question_options") else q.question_options
+                ),
+            }
+            for q in survey.questions
+        }
+
     @property
-    def sub_dicts(self) -> dict[str, dict]:
-        """Return a dictionary where keys are strings for each of the main class attributes/objects."""
-        if self.agent.name is None:
-            agent_name = agent_namer(self.agent)
+    def agent(self) -> "Agent":
+        """Return the Agent object."""
+        return self.data["agent"]
+
+    @property
+    def scenario(self) -> "Scenario":
+        """Return the Scenario object."""
+        return self.data["scenario"]
+
+    @property
+    def model(self) -> "LanguageModel":
+        """Return the LanguageModel object."""
+        return self.data["model"]
+
+    @property
+    def answer(self) -> dict[QuestionName, AnswerValue]:
+        """Return the answers."""
+        return self.data["answer"]
+
+    @staticmethod
+    def _create_agent_sub_dict(agent) -> dict:
+        """Create a dictionary of agent details"""
+        if agent.name is None:
+            agent_name = agent_namer(agent)
         else:
-            agent_name = self.agent.name
-
-        # comments_dict = {k: v for k, v in self.answer.items() if k.endswith("_comment")}
-        question_text_dict = {}
-        question_options_dict = {}
-        question_type_dict = {}
-        for key, _ in self.answer.items():
-            if key in self.question_to_attributes:
-                # You might be tempted to just use the naked key
-                # but this is a bad idea because it pollutes the namespace
-                question_text_dict[
-                    key + "_question_text"
-                ] = self.question_to_attributes[key]["question_text"]
-                question_options_dict[
-                    key + "_question_options"
-                ] = self.question_to_attributes[key]["question_options"]
-                question_type_dict[
-                    key + "_question_type"
-                ] = self.question_to_attributes[key]["question_type"]
+            agent_name = agent.name
 
         return {
-            "agent": self.agent.traits
+            "agent": agent.traits
             | {"agent_name": agent_name}
-            | {"agent_instruction": self.agent.instruction},
-            "scenario": self.scenario,
-            "model": self.model.parameters | {"model": self.model.model},
-            "answer": self.answer,
-            "prompt": self.prompt,
-            "raw_model_response": self.raw_model_response,
-            "iteration": {"iteration": self.iteration},
-            "question_text": question_text_dict,
-            "question_options": question_options_dict,
-            "question_type": question_type_dict,
-            "comment": self.comments_dict,
-            "generated_tokens": self.generated_tokens,
+            | {"agent_instruction": agent.instruction},
+        }
+
+    @staticmethod
+    def _create_model_sub_dict(model) -> dict:
+        return {
+            "model": model.parameters | {"model": model.model},
+        }
+
+    @staticmethod
+    def _iteration_sub_dict(iteration) -> dict:
+        return {
+            "iteration": {"iteration": iteration},
+        }
+
+    def _construct_sub_dicts(self) -> dict[str, dict]:
+        """Construct a dictionary of sub-dictionaries for the Result object."""
+        sub_dicts_needing_new_keys = {
+            "question_text": {},
+            "question_options": {},
+            "question_type": {},
+        }
+
+        for question_name in self.data["answer"]:
+            if question_name in self.question_to_attributes:
+                for dictionary_name in sub_dicts_needing_new_keys:
+                    new_key = question_name + "_" + dictionary_name
+                    sub_dicts_needing_new_keys[dictionary_name][new_key] = (
+                        self.question_to_attributes[question_name][dictionary_name]
+                    )
+
+        new_cache_dict = {
+            f"{k}_cache_used": v for k, v in self.data["cache_used_dict"].items()
         }
 
-    def check_expression(self, expression) -> None:
+        d = {
+            **self._create_agent_sub_dict(self.data["agent"]),
+            **self._create_model_sub_dict(self.data["model"]),
+            **self._iteration_sub_dict(self.data["iteration"]),
+            "scenario": self.data["scenario"],
+            "answer": self.data["answer"],
+            "prompt": self.data["prompt"],
+            "comment": self.data["comments_dict"],
+            "generated_tokens": self.data["generated_tokens"],
+            "raw_model_response": self.data["raw_model_response"],
+            "question_text": sub_dicts_needing_new_keys["question_text"],
+            "question_options": sub_dicts_needing_new_keys["question_options"],
+            "question_type": sub_dicts_needing_new_keys["question_type"],
+            "cache_used": new_cache_dict,
+        }
+        if hasattr(self, "indices") and self.indices is not None:
+            d["agent"].update({"agent_index": self.indices["agent"]})
+            d["scenario"].update({"scenario_index": self.indices["scenario"]})
+            d["model"].update({"model_index": self.indices["model"]})
+        return d
+
+    @property
+    def sub_dicts(self) -> dict[str, dict]:
+        """Return a dictionary where keys are strings for each of the main class attributes/objects."""
+        if self._sub_dicts is None:
+            self._sub_dicts = self._construct_sub_dicts()
+        return self._sub_dicts
+
+    def check_expression(self, expression: str) -> None:
         for key in self.problem_keys:
             if key in expression and not key + "." in expression:
                 raise ValueError(
@@ -205,11 +222,13 @@ class Result(Base, UserDict):
         raise NotImplementedError
 
     @property
-    def problem_keys(self):
+    def problem_keys(self) -> list[str]:
         """Return a list of keys that are problematic."""
         return self._problem_keys
 
-    def _compute_combined_dict_and_problem_keys(self) -> None:
+    def _compute_combined_dict_and_problem_keys(
+        self,
+    ) -> tuple[dict[str, Any], list[str]]:
         combined = {}
         problem_keys = []
         for key, sub_dict in self.sub_dicts.items():
@@ -222,8 +241,7 @@ class Result(Base, UserDict):
             combined.update({key: sub_dict})
             # I *think* this allows us to do do things like "answer.how_feelling" i.e., that the evaluator can use
             # dot notation to access the subdicts.
-        self._combined_dict = combined
-        self._problem_keys = problem_keys
+        return combined, problem_keys
 
     @property
     def combined_dict(self) -> dict[str, Any]:
@@ -234,11 +252,14 @@ class Result(Base, UserDict):
         'OK'
         """
         if self._combined_dict is None or self._problem_keys is None:
-            self._compute_combined_dict_and_problem_keys()
+            (
+                self._combined_dict,
+                self._problem_keys,
+            ) = self._compute_combined_dict_and_problem_keys()
         return self._combined_dict
 
     @property
-    def problem_keys(self):
+    def problem_keys(self) -> list[str]:
         """Return a list of keys that are problematic."""
         if self._combined_dict is None or self._problem_keys is None:
             self._compute_combined_dict_and_problem_keys()
@@ -278,7 +299,6 @@ class Result(Base, UserDict):
                     )
                     problem_keys.append((key, data_type))
                     key = f"{key}_{data_type}"
-                    # raise ValueError(f"Key '{key}' is already in the dictionary")
                 d[key] = data_type
 
         for key, data_type in problem_keys:
@@ -287,37 +307,16 @@ class Result(Base, UserDict):
             ].pop(key)
         return d
 
-    def rows(self, index) -> tuple[int, str, str, str]:
-        """Return a generator of rows for the Result object."""
-        for data_type, subdict in self.sub_dicts.items():
-            for key, value in subdict.items():
-                yield (index, data_type, key, str(value))
-
-    def leaves(self):
-        leaves = []
-        for question_name, answer in self.answer.items():
-            if not question_name.endswith("_comment"):
-                leaves.append(
-                    {
-                        "question": f"({question_name}): "
-                        + str(
-                            self.question_to_attributes[question_name]["question_text"]
-                        ),
-                        "answer": answer,
-                        "comment": self.answer.get(question_name + "_comment", ""),
-                        "scenario": repr(self.scenario),
-                        "agent": repr(self.agent),
-                        "model": repr(self.model),
-                        "iteration": self.iteration,
-                    }
-                )
-        return leaves
-
-    ###############
-    # Useful
-    ###############
     def copy(self) -> Result:
-        """Return a copy of the Result object."""
+        """Return a copy of the Result object.
+
+        >>> r = Result.example()
+        >>> r2 = r.copy()
+        >>> r == r2
+        True
+        >>> id(r) == id(r2)
+        False
+        """
         return Result.from_dict(self.to_dict())
 
     def __eq__(self, other) -> bool:
@@ -328,17 +327,16 @@ class Result(Base, UserDict):
         True
 
         """
-        return self.to_dict() == other.to_dict()
+        return hash(self) == hash(other)
 
-    ###############
-    # Serialization
-    ###############
-    def to_dict(self, add_edsl_version=True) -> dict[str, Any]:
+    def to_dict(
+        self, add_edsl_version: bool = True, include_cache_info: bool = False
+    ) -> dict[str, Any]:
         """Return a dictionary representation of the Result object.
 
         >>> r = Result.example()
         >>> r.to_dict()['scenario']
-        {'period': 'morning', 'edsl_version': '...', 'edsl_class_name': 'Scenario'}
+        {'period': 'morning', 'scenario_index': 0, 'edsl_version': '...', 'edsl_class_name': 'Scenario'}
         """
 
         def convert_value(value, add_edsl_version=True):
@@ -366,21 +364,26 @@ class Result(Base, UserDict):
             d["edsl_version"] = __version__
             d["edsl_class_name"] = "Result"
 
+        if include_cache_info:
+            d["cache_used_dict"] = self.data["cache_used_dict"]
+        else:
+            d.pop("cache_used_dict", None)
+
         return d
 
     def __hash__(self):
         """Return a hash of the Result object."""
         from edsl.utilities.utilities import dict_hash
 
-        return dict_hash(self.to_dict(add_edsl_version=False))
+        return dict_hash(self.to_dict(add_edsl_version=False, include_cache_info=False))
 
     @classmethod
     @remove_edsl_version
     def from_dict(self, json_dict: dict) -> Result:
         """Return a Result object from a dictionary representation."""
 
-        from edsl import Agent
-        from edsl import Scenario
+        from edsl.agents.Agent import Agent
+        from edsl.scenarios.Scenario import Scenario
         from edsl.language_models.LanguageModel import LanguageModel
         from edsl.prompts.Prompt import Prompt
 
@@ -402,51 +405,34 @@ class Result(Base, UserDict):
             question_to_attributes=json_dict.get("question_to_attributes", None),
             generated_tokens=json_dict.get("generated_tokens", {}),
             comments_dict=json_dict.get("comments_dict", {}),
+            cache_used_dict=json_dict.get("cache_used_dict", {}),
         )
         return result
 
-    def rich_print(self) -> None:
-        """Display an object as a table."""
-        # from edsl.utilities import print_dict_with_rich
-        from rich import print
-        from rich.table import Table
-
-        table = Table(title="Result")
-        table.add_column("Attribute", style="bold")
-        table.add_column("Value")
-
-        to_display = self.__dict__.copy()
-        data = to_display.pop("data", None)
-        for attr_name, attr_value in to_display.items():
-            if hasattr(attr_value, "rich_print"):
-                table.add_row(attr_name, attr_value.rich_print())
-            elif isinstance(attr_value, dict):
-                a = PromptDict(attr_value)
-                table.add_row(attr_name, a.rich_print())
-            else:
-                table.add_row(attr_name, repr(attr_value))
-        return table
-
     def __repr__(self):
         """Return a string representation of the Result object."""
-        return f"Result(agent={repr(self.agent)}, scenario={repr(self.scenario)}, model={repr(self.model)}, iteration={self.iteration}, answer={repr(self.answer)}, prompt={repr(self.prompt)})"
+        params = ", ".join(f"{key}={repr(value)}" for key, value in self.data.items())
+        return f"{self.__class__.__name__}({params})"
 
     @classmethod
     def example(cls):
-        """Return an example Result object."""
+        """Return an example Result object.
+
+        >>> Result.example()
+        Result(...)
+
+        """
         from edsl.results.Results import Results
 
         return Results.example()[0]
 
-    def score(self, scoring_function: Callable) -> Any:
+    def score(self, scoring_function: Callable) -> Union[int, float]:
         """Score the result using a passed-in scoring function.
 
         >>> def f(status): return 1 if status == 'Joyful' else 0
         >>> Result.example().score(f)
         1
         """
-        import inspect
-
         signature = inspect.signature(scoring_function)
         params = {}
         for k, v in signature.parameters.items():
@@ -458,6 +444,112 @@ class Result(Base, UserDict):
                 raise ValueError(f"Parameter {k} not found in Result object")
         return scoring_function(**params)
 
+    @classmethod
+    def from_interview(
+        cls, interview, extracted_answers, model_response_objects
+    ) -> Result:
+        """Return a Result object from an interview dictionary."""
+
+        def get_question_results(
+            model_response_objects,
+        ) -> dict[str, "EDSLResultObjectInput"]:
+            """Maps the question name to the EDSLResultObjectInput."""
+            question_results = {}
+            for result in model_response_objects:
+                question_results[result.question_name] = result
+            return question_results
+
+        def get_generated_tokens_dict(answer_key_names) -> dict[str, str]:
+            generated_tokens_dict = {
+                k + "_generated_tokens": question_results[k].generated_tokens
+                for k in answer_key_names
+            }
+            return generated_tokens_dict
+
+        def get_comments_dict(answer_key_names) -> dict[str, str]:
+            comments_dict = {
+                k + "_comment": question_results[k].comment for k in answer_key_names
+            }
+            return comments_dict
+
+        def get_question_name_to_prompts(
+            model_response_objects,
+        ) -> dict[str, dict[str, str]]:
+            question_name_to_prompts = dict({})
+            for result in model_response_objects:
+                question_name = result.question_name
+                question_name_to_prompts[question_name] = {
+                    "user_prompt": result.prompts["user_prompt"],
+                    "system_prompt": result.prompts["system_prompt"],
+                }
+            return question_name_to_prompts
+
+        def get_prompt_dictionary(answer_key_names, question_name_to_prompts):
+            prompt_dictionary = {}
+            for answer_key_name in answer_key_names:
+                prompt_dictionary[answer_key_name + "_user_prompt"] = (
+                    question_name_to_prompts[answer_key_name]["user_prompt"]
+                )
+                prompt_dictionary[answer_key_name + "_system_prompt"] = (
+                    question_name_to_prompts[answer_key_name]["system_prompt"]
+                )
+            return prompt_dictionary
+
+        def get_raw_model_results_and_cache_used_dictionary(model_response_objects):
+            raw_model_results_dictionary = {}
+            cache_used_dictionary = {}
+            for result in model_response_objects:
+                question_name = result.question_name
+                raw_model_results_dictionary[question_name + "_raw_model_response"] = (
+                    result.raw_model_response
+                )
+                raw_model_results_dictionary[question_name + "_cost"] = result.cost
+                one_use_buys = (
+                    "NA"
+                    if isinstance(result.cost, str)
+                    or result.cost == 0
+                    or result.cost is None
+                    else 1.0 / result.cost
+                )
+                raw_model_results_dictionary[question_name + "_one_usd_buys"] = (
+                    one_use_buys
+                )
+                cache_used_dictionary[question_name] = result.cache_used
+
+            return raw_model_results_dictionary, cache_used_dictionary
+
+        question_results = get_question_results(model_response_objects)
+        answer_key_names = list(question_results.keys())
+        generated_tokens_dict = get_generated_tokens_dict(answer_key_names)
+        comments_dict = get_comments_dict(answer_key_names)
+        answer_dict = {k: extracted_answers[k] for k in answer_key_names}
+
+        question_name_to_prompts = get_question_name_to_prompts(model_response_objects)
+        prompt_dictionary = get_prompt_dictionary(
+            answer_key_names, question_name_to_prompts
+        )
+        raw_model_results_dictionary, cache_used_dictionary = (
+            get_raw_model_results_and_cache_used_dictionary(model_response_objects)
+        )
+
+        result = cls(
+            agent=interview.agent,
+            scenario=interview.scenario,
+            model=interview.model,
+            iteration=interview.iteration,
+            # Computed objects
+            answer=answer_dict,
+            prompt=prompt_dictionary,
+            raw_model_response=raw_model_results_dictionary,
+            survey=interview.survey,
+            generated_tokens=generated_tokens_dict,
+            comments_dict=comments_dict,
+            cache_used_dict=cache_used_dictionary,
+            indices=interview.indices,
+        )
+        result.interview_hash = interview.initial_hash
+        return result
+
 
 if __name__ == "__main__":
     import doctest