edsl 0.1.32__py3-none-any.whl → 0.1.33__py3-none-any.whl

edsl/jobs/interviews/InterviewExceptionEntry.py

@@ -2,24 +2,62 @@ import traceback
 import datetime
 import time
 from collections import UserDict
-
-# traceback=traceback.format_exc(),
-# traceback = frame_summary_to_dict(traceback.extract_tb(e.__traceback__))
-# traceback = [frame_summary_to_dict(f) for f in traceback.extract_tb(e.__traceback__)]
+from edsl.jobs.FailedQuestion import FailedQuestion
 
 
 class InterviewExceptionEntry:
-    """Class to record an exception that occurred during the interview.
-
-    >>> entry = InterviewExceptionEntry.example()
-    >>> entry.to_dict()['exception']
-    "ValueError('An error occurred.')"
-    """
-
-    def __init__(self, exception: Exception, traceback_format="html"):
+    """Class to record an exception that occurred during the interview."""
+
+    def __init__(
+        self,
+        *,
+        exception: Exception,
+        # failed_question: FailedQuestion,
+        invigilator: "Invigilator",
+        traceback_format="text",
+        answers=None,
+    ):
         self.time = datetime.datetime.now().isoformat()
         self.exception = exception
+        # self.failed_question = failed_question
+        self.invigilator = invigilator
         self.traceback_format = traceback_format
+        self.answers = answers
+
+    @property
+    def question_type(self):
+        # return self.failed_question.question.question_type
+        return self.invigilator.question.question_type
+
+    @property
+    def name(self):
+        return repr(self.exception)
+
+    @property
+    def rendered_prompts(self):
+        return self.invigilator.get_prompts()
+
+    @property
+    def key_sequence(self):
+        return self.invigilator.model.key_sequence
+
+    @property
+    def generated_token_string(self):
+        # return "POO"
+        if self.invigilator.raw_model_response is None:
+            return "No raw model response available."
+        else:
+            return self.invigilator.model.get_generated_token_string(
+                self.invigilator.raw_model_response
+            )
+
+    @property
+    def raw_model_response(self):
+        import json
+
+        if self.invigilator.raw_model_response is None:
+            return "No raw model response available."
+        return json.dumps(self.invigilator.raw_model_response, indent=2)
 
     def __getitem__(self, key):
         # Support dict-like access obj['a']
@@ -27,11 +65,37 @@ class InterviewExceptionEntry:
 
     @classmethod
     def example(cls):
-        try:
-            raise ValueError("An error occurred.")
-        except Exception as e:
-            entry = InterviewExceptionEntry(e)
-        return entry
+        from edsl import QuestionFreeText
+        from edsl.language_models import LanguageModel
+
+        m = LanguageModel.example(test_model=True)
+        q = QuestionFreeText.example(exception_to_throw=ValueError)
+        results = q.by(m).run(
+            skip_retry=True, print_exceptions=False, raise_validation_errors=True
+        )
+        return results.task_history.exceptions[0]["how_are_you"][0]
+
+    @property
+    def code_to_reproduce(self):
+        return self.code(run=False)
+
+    def code(self, run=True):
+        lines = []
+        lines.append("from edsl import Question, Model, Scenario, Agent")
+
+        lines.append(f"q = {repr(self.invigilator.question)}")
+        lines.append(f"scenario = {repr(self.invigilator.scenario)}")
+        lines.append(f"agent = {repr(self.invigilator.agent)}")
+        lines.append(f"m = Model('{self.invigilator.model.model}')")
+        lines.append("results = q.by(m).by(agent).by(scenario).run()")
+        code_str = "\n".join(lines)
+
+        if run:
+            # Create a new namespace to avoid polluting the global namespace
+            namespace = {}
+            exec(code_str, namespace)
+            return namespace["results"]
+        return code_str
 
     @property
     def traceback(self):
@@ -78,13 +142,15 @@ class InterviewExceptionEntry:
 
         >>> entry = InterviewExceptionEntry.example()
         >>> entry.to_dict()['exception']
-        "ValueError('An error occurred.')"
+        ValueError()
 
         """
         return {
-            "exception": repr(self.exception),
+            "exception": self.exception,
             "time": self.time,
             "traceback": self.traceback,
+            # "failed_question": self.failed_question.to_dict(),
+            "invigilator": self.invigilator.to_dict(),
         }
 
     def push(self):

edsl/jobs/runners/JobsRunnerAsyncio.py

@@ -1,42 +1,27 @@
 from __future__ import annotations
 import time
+import math
 import asyncio
-import time
+import functools
+import threading
+from typing import Coroutine, List, AsyncGenerator, Optional, Union, Generator
 from contextlib import contextmanager
+from collections import UserList
 
-from typing import Coroutine, List, AsyncGenerator, Optional, Union
+from edsl.results.Results import Results
+from rich.live import Live
+from rich.console import Console
 
 from edsl import shared_globals
 from edsl.jobs.interviews.Interview import Interview
-from edsl.jobs.runners.JobsRunnerStatusMixin import JobsRunnerStatusMixin
+from edsl.jobs.runners.JobsRunnerStatus import JobsRunnerStatus
+
 from edsl.jobs.tasks.TaskHistory import TaskHistory
 from edsl.jobs.buckets.BucketCollection import BucketCollection
 from edsl.utilities.decorators import jupyter_nb_handler
-
-import time
-import functools
-
-
-def cache_with_timeout(timeout):
-    def decorator(func):
-        cached_result = {}
-        last_computation_time = [0]  # Using list to store mutable value
-
-        @functools.wraps(func)
-        def wrapper(*args, **kwargs):
-            current_time = time.time()
-            if (current_time - last_computation_time[0]) >= timeout:
-                cached_result["value"] = func(*args, **kwargs)
-                last_computation_time[0] = current_time
-            return cached_result["value"]
-
-        return wrapper
-
-    return decorator
-
-
-# from queue import Queue
-from collections import UserList
+from edsl.data.Cache import Cache
+from edsl.results.Result import Result
+from edsl.results.Results import Results
 
 
 class StatusTracker(UserList):
@@ -48,99 +33,87 @@ class StatusTracker(UserList):
         return print(f"Completed: {len(self.data)} of {self.total_tasks}", end="\r")
 
 
-class JobsRunnerAsyncio(JobsRunnerStatusMixin):
+class JobsRunnerAsyncio:
     """A class for running a collection of interviews asynchronously.
 
     It gets instaniated from a Jobs object.
     The Jobs object is a collection of interviews that are to be run.
     """
 
-    def __init__(self, jobs: Jobs):
+    def __init__(self, jobs: "Jobs"):
         self.jobs = jobs
-        # this creates the interviews, which can take a while
         self.interviews: List["Interview"] = jobs.interviews()
         self.bucket_collection: "BucketCollection" = jobs.bucket_collection
         self.total_interviews: List["Interview"] = []
 
+        # self.jobs_runner_status = JobsRunnerStatus(self, n=1)
+
     async def run_async_generator(
         self,
         cache: "Cache",
         n: int = 1,
-        debug: bool = False,
         stop_on_exception: bool = False,
-        sidecar_model: "LanguageModel" = None,
+        sidecar_model: Optional["LanguageModel"] = None,
         total_interviews: Optional[List["Interview"]] = None,
+        raise_validation_errors: bool = False,
     ) -> AsyncGenerator["Result", None]:
         """Creates the tasks, runs them asynchronously, and returns the results as a Results object.
 
         Completed tasks are yielded as they are completed.
 
         :param n: how many times to run each interview
-        :param debug:
         :param stop_on_exception: Whether to stop the interview if an exception is raised
         :param sidecar_model: a language model to use in addition to the interview's model
         :param total_interviews: A list of interviews to run can be provided instead.
+        :param raise_validation_errors: Whether to raise validation errors
         """
         tasks = []
-        if total_interviews:
+        if total_interviews:  # was already passed in total interviews
             self.total_interviews = total_interviews
         else:
-            self._populate_total_interviews(
-                n=n
+            self.total_interviews = list(
+                self._populate_total_interviews(n=n)
             )  # Populate self.total_interviews before creating tasks
 
         for interview in self.total_interviews:
             interviewing_task = self._build_interview_task(
                 interview=interview,
-                debug=debug,
                 stop_on_exception=stop_on_exception,
                 sidecar_model=sidecar_model,
+                raise_validation_errors=raise_validation_errors,
             )
             tasks.append(asyncio.create_task(interviewing_task))
 
         for task in asyncio.as_completed(tasks):
             result = await task
+            self.jobs_runner_status.add_completed_interview(result)
             yield result
 
-    def _populate_total_interviews(self, n: int = 1) -> None:
+    def _populate_total_interviews(
+        self, n: int = 1
+    ) -> Generator["Interview", None, None]:
         """Populates self.total_interviews with n copies of each interview.
 
         :param n: how many times to run each interview.
         """
-        # TODO: Why not return a list of interviews instead of modifying the object?
-
-        self.total_interviews = []
         for interview in self.interviews:
             for iteration in range(n):
                 if iteration > 0:
-                    new_interview = interview.duplicate(
-                        iteration=iteration, cache=self.cache
-                    )
-                    self.total_interviews.append(new_interview)
+                    yield interview.duplicate(iteration=iteration, cache=self.cache)
                 else:
-                    interview.cache = (
-                        self.cache
-                    )  # set the cache for the first interview
-                    self.total_interviews.append(interview)
-
-    async def run_async(self, cache=None, n=1) -> Results:
-        from edsl.results.Results import Results
+                    interview.cache = self.cache
+                    yield interview
 
-        # breakpoint()
-        # tracker = StatusTracker(total_tasks=len(self.interviews))
-
-        if cache is None:
-            self.cache = Cache()
-        else:
-            self.cache = cache
+    async def run_async(self, cache: Optional["Cache"] = None, n: int = 1) -> Results:
+        """Used for some other modules that have a non-standard way of running interviews."""
+        self.jobs_runner_status = JobsRunnerStatus(self, n=n)
+        self.cache = Cache() if cache is None else cache
         data = []
         async for result in self.run_async_generator(cache=self.cache, n=n):
             data.append(result)
         return Results(survey=self.jobs.survey, data=data)
 
     def simple_run(self):
-        from edsl.results.Results import Results
-
         data = asyncio.run(self.run_async())
         return Results(survey=self.jobs.survey, data=data)
 
@@ -148,14 +121,13 @@ class JobsRunnerAsyncio(JobsRunnerStatusMixin):
         self,
         *,
         interview: Interview,
-        debug: bool,
         stop_on_exception: bool = False,
-        sidecar_model: Optional[LanguageModel] = None,
-    ) -> Result:
+        sidecar_model: Optional["LanguageModel"] = None,
+        raise_validation_errors: bool = False,
+    ) -> "Result":
         """Conducts an interview and returns the result.
 
         :param interview: the interview to conduct
-        :param debug: prints debug messages
         :param stop_on_exception: stops the interview if an exception is raised
         :param sidecar_model: a language model to use in addition to the interview's model
         """
@@ -164,24 +136,37 @@ class JobsRunnerAsyncio(JobsRunnerStatusMixin):
 
         # get the results of the interview
         answer, valid_results = await interview.async_conduct_interview(
-            debug=debug,
             model_buckets=model_buckets,
             stop_on_exception=stop_on_exception,
             sidecar_model=sidecar_model,
+            raise_validation_errors=raise_validation_errors,
         )
 
-        # we should have a valid result for each question
-        answer_key_names = {k for k in set(answer.keys()) if not k.endswith("_comment")}
+        question_results = {}
+        for result in valid_results:
+            question_results[result.question_name] = result
 
+        answer_key_names = list(question_results.keys())
+
+        generated_tokens_dict = {
+            k + "_generated_tokens": question_results[k].generated_tokens
+            for k in answer_key_names
+        }
+        comments_dict = {
+            k + "_comment": question_results[k].comment for k in answer_key_names
+        }
+
+        # we should have a valid result for each question
+        answer_dict = {k: answer[k] for k in answer_key_names}
         assert len(valid_results) == len(answer_key_names)
 
         # TODO: move this down into Interview
         question_name_to_prompts = dict({})
         for result in valid_results:
-            question_name = result["question_name"]
+            question_name = result.question_name
             question_name_to_prompts[question_name] = {
-                "user_prompt": result["prompts"]["user_prompt"],
-                "system_prompt": result["prompts"]["system_prompt"],
+                "user_prompt": result.prompts["user_prompt"],
+                "system_prompt": result.prompts["system_prompt"],
             }
 
         prompt_dictionary = {}
@@ -195,22 +180,31 @@ class JobsRunnerAsyncio(JobsRunnerStatusMixin):
 
         raw_model_results_dictionary = {}
         for result in valid_results:
-            question_name = result["question_name"]
+            question_name = result.question_name
             raw_model_results_dictionary[
                 question_name + "_raw_model_response"
-            ] = result["raw_model_response"]
-
-        from edsl.results.Result import Result
+            ] = 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
 
         result = Result(
             agent=interview.agent,
             scenario=interview.scenario,
             model=interview.model,
             iteration=interview.iteration,
-            answer=answer,
+            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,
         )
         result.interview_hash = hash(interview)
 
@@ -220,132 +214,109 @@ class JobsRunnerAsyncio(JobsRunnerStatusMixin):
     def elapsed_time(self):
         return time.monotonic() - self.start_time
 
+    def process_results(
+        self, raw_results: Results, cache: Cache, print_exceptions: bool
+    ):
+        interview_lookup = {
+            hash(interview): index
+            for index, interview in enumerate(self.total_interviews)
+        }
+        interview_hashes = list(interview_lookup.keys())
+
+        results = Results(
+            survey=self.jobs.survey,
+            data=sorted(
+                raw_results, key=lambda x: interview_hashes.index(x.interview_hash)
+            ),
+        )
+        results.cache = cache
+        results.task_history = TaskHistory(
+            self.total_interviews, include_traceback=False
+        )
+        results.has_unfixed_exceptions = results.task_history.has_unfixed_exceptions
+        results.bucket_collection = self.bucket_collection
+
+        if results.has_unfixed_exceptions and print_exceptions:
+            from edsl.scenarios.FileStore import HTMLFileStore
+            from edsl.config import CONFIG
+            from edsl.coop.coop import Coop
+
+            msg = f"Exceptions were raised in {len(results.task_history.indices)} out of {len(self.total_interviews)} interviews.\n"
+
+            if len(results.task_history.indices) > 5:
+                msg += f"Exceptions were raised in the following interviews: {results.task_history.indices}.\n"
+
+            print(msg)
+            # this is where exceptions are opening up
+            filepath = results.task_history.html(
+                cta="Open report to see details.",
+                open_in_browser=True,
+                return_link=True,
+            )
+
+            try:
+                coop = Coop()
+                user_edsl_settings = coop.edsl_settings
+                remote_logging = user_edsl_settings["remote_logging"]
+            except Exception as e:
+                print(e)
+                remote_logging = False
+            if remote_logging:
+                filestore = HTMLFileStore(filepath)
+                coop_details = filestore.push(description="Error report")
+                print(coop_details)
+
+            print("Also see: https://docs.expectedparrot.com/en/latest/exceptions.html")
+
+        return results
+
     @jupyter_nb_handler
     async def run(
         self,
         cache: Union[Cache, False, None],
         n: int = 1,
-        debug: bool = False,
         stop_on_exception: bool = False,
         progress_bar: bool = False,
         sidecar_model: Optional[LanguageModel] = None,
         print_exceptions: bool = True,
+        raise_validation_errors: bool = False,
     ) -> "Coroutine":
         """Runs a collection of interviews, handling both async and sync contexts."""
-        from rich.console import Console
 
-        console = Console()
         self.results = []
         self.start_time = time.monotonic()
         self.completed = False
         self.cache = cache
         self.sidecar_model = sidecar_model
 
-        from edsl.results.Results import Results
-        from rich.live import Live
-        from rich.console import Console
-
-        @cache_with_timeout(1)
-        def generate_table():
-            return self.status_table(self.results, self.elapsed_time)
+        self.jobs_runner_status = JobsRunnerStatus(self, n=n)
 
-        async def process_results(cache, progress_bar_context=None):
+        async def process_results(cache):
             """Processes results from interviews."""
             async for result in self.run_async_generator(
                 n=n,
-                debug=debug,
                 stop_on_exception=stop_on_exception,
                 cache=cache,
                 sidecar_model=sidecar_model,
+                raise_validation_errors=raise_validation_errors,
             ):
                 self.results.append(result)
-                if progress_bar_context:
-                    progress_bar_context.update(generate_table())
-                self.completed = True
-
-        async def update_progress_bar(progress_bar_context):
-            """Updates the progress bar at fixed intervals."""
-            if progress_bar_context is None:
-                return
-
-            while True:
-                progress_bar_context.update(generate_table())
-                await asyncio.sleep(0.1)  # Update interval
-                if self.completed:
-                    break
-
-        @contextmanager
-        def conditional_context(condition, context_manager):
-            if condition:
-                with context_manager as cm:
-                    yield cm
-            else:
-                yield
-
-        with conditional_context(
-            progress_bar, Live(generate_table(), console=console, refresh_per_second=1)
-        ) as progress_bar_context:
-            with cache as c:
-                progress_task = asyncio.create_task(
-                    update_progress_bar(progress_bar_context)
-                )
-
-                try:
-                    await asyncio.gather(
-                        progress_task,
-                        process_results(
-                            cache=c, progress_bar_context=progress_bar_context
-                        ),
-                    )
-                except asyncio.CancelledError:
-                    pass
-                finally:
-                    progress_task.cancel()  # Cancel the progress_task when process_results is done
-                    await progress_task
-
-                    await asyncio.sleep(1)  # short delay to show the final status
-
-                    if progress_bar_context:
-                        progress_bar_context.update(generate_table())
-
-        # puts results in the same order as the total interviews
-        interview_hashes = [hash(interview) for interview in self.total_interviews]
-        self.results = sorted(
-            self.results, key=lambda x: interview_hashes.index(x.interview_hash)
-        )
+            self.completed = True
 
-        results = Results(survey=self.jobs.survey, data=self.results)
-        task_history = TaskHistory(self.total_interviews, include_traceback=False)
-        results.task_history = task_history
-
-        results.has_exceptions = task_history.has_exceptions
-
-        if results.has_exceptions:
-            # put the failed interviews in the results object as a list
-            failed_interviews = [
-                interview.duplicate(
-                    iteration=interview.iteration, cache=interview.cache
-                )
-                for interview in self.total_interviews
-                if interview.has_exceptions
-            ]
-            from edsl.jobs.Jobs import Jobs
-
-            results.failed_jobs = Jobs.from_interviews(
-                [interview for interview in failed_interviews]
-            )
-            if print_exceptions:
-                msg = f"Exceptions were raised in {len(results.task_history.indices)} out of {len(self.total_interviews)} interviews.\n"
+        def run_progress_bar():
+            """Runs the progress bar in a separate thread."""
+            self.jobs_runner_status.update_progress()
 
-                if len(results.task_history.indices) > 5:
-                    msg += f"Exceptions were raised in the following interviews: {results.task_history.indices}.\n"
+        if progress_bar:
+            progress_thread = threading.Thread(target=run_progress_bar)
+            progress_thread.start()
 
-                shared_globals["edsl_runner_exceptions"] = task_history
-                print(msg)
-                # this is where exceptions are opening up
-                task_history.html(cta="Open report to see details.")
-                print(
-                    "Also see: https://docs.expectedparrot.com/en/latest/exceptions.html"
-                )
+        with cache as c:
+            await process_results(cache=c)
 
-        return results
+        if progress_bar:
+            progress_thread.join()
+
+        return self.process_results(
+            raw_results=self.results, cache=cache, print_exceptions=print_exceptions
+        )