"""
The `learning` module is designed to facilitate the collection and storage of user feedback on the outputs generated by the GPT Engineer tool. It provides mechanisms for obtaining user consent, capturing user reviews, and storing this information for future analysis and enhancement of the tool's performance.
Classes
-------
Review : dataclass
Represents a user's review of the generated code, including whether it ran, was perfect, was useful, and any additional comments.
Learning : dataclass
Encapsulates the metadata and feedback collected during a session of using the GPT Engineer tool, including the prompt, model, temperature, configuration, logs, session identifier, user review, and timestamp.
Functions
---------
human_review_input() -> Optional[Review]
Interactively gathers feedback from the user regarding the performance of generated code and returns a Review instance.
check_collection_consent() -> bool
Checks if the user has previously given consent to store their data and, if not, asks for it.
ask_collection_consent() -> bool
Prompts the user for consent to store their data for the purpose of improving GPT Engineer.
extract_learning(prompt: Prompt, model: str, temperature: float, config: Tuple[str, ...], memory: DiskMemory, review: Review) -> Learning
Extracts feedback and session details to create a Learning instance based on the provided parameters.
get_session() -> str
Retrieves a unique identifier for the current user session, creating one if it does not exist.
Constants
---------
TERM_CHOICES : tuple
Terminal color choices for user interactive prompts, formatted with termcolor for readability.
"""
import json
import random
import tempfile
from dataclasses import dataclass, field
from datetime import datetime
from pathlib import Path
from typing import Optional, Tuple
from dataclasses_json import dataclass_json
from termcolor import colored
from gpt_engineer.core.default.disk_memory import DiskMemory
from gpt_engineer.core.prompt import Prompt
@dataclass_json
@dataclass
class Review:
"""
A dataclass that represents a user's review of the generated code.
Attributes
----------
ran : Optional[bool]
Indicates whether the generated code ran without errors.
perfect : Optional[bool]
Indicates whether the generated code met all the user's requirements.
works : Optional[bool]
Indicates whether the generated code was useful, even if not perfect.
comments : str
Any additional comments provided by the user.
raw : str
A raw string representation of the user's responses.
"""
ran: Optional[bool]
perfect: Optional[bool]
works: Optional[bool]
comments: str
raw: str
@dataclass_json
@dataclass
class Learning:
"""
A dataclass that encapsulates the learning data collected during a GPT Engineer session.
Attributes
----------
prompt : str
A JSON string representing the prompt provided to GPT Engineer.
model : str
The name of the model used during the session.
temperature : float
The temperature setting used for the model's responses.
config : str
A JSON string representing the configuration settings for the session.
logs : str
A JSON string representing the logs of the session.
session : str
A unique identifier for the user session.
review : Optional[Review]
The user's review of the generated code.
timestamp : str
The UTC timestamp when the learning data was created.
version : str
The version of the learning data schema.
"""
prompt: str
model: str
temperature: float
config: str
logs: str
session: str
review: Optional[Review]
timestamp: str = field(default_factory=lambda: datetime.utcnow().isoformat())
version: str = "0.3"
TERM_CHOICES = (
colored("y", "green")
+ "/"
+ colored("n", "red")
+ "/"
+ colored("u", "yellow")
+ "(ncertain): "
)
[docs]
def check_collection_consent() -> bool:
"""
Checks if the user has previously given consent to store their data for feedback collection.
This function looks for a file that stores the user's consent status. If the file exists and contains 'true', consent is assumed. If the file does not exist or does not contain 'true', the function will prompt the user for consent.
Returns
-------
bool
True if the user has given consent, False otherwise.
"""
path = Path(".gpte_consent")
if path.exists() and path.read_text() == "true":
return True
else:
return ask_collection_consent()
[docs]
def ask_collection_consent() -> bool:
"""
Asks the user for their consent to store their data for the purpose of improving the GPT Engineer tool.
The user's response is recorded in a file for future reference. If the user consents, the function will write 'true' to the file. If the user does not consent, no data will be collected, and the function will not modify the file.
Returns
-------
bool
True if the user consents, False otherwise.
"""
answer = input(
"Is it ok if we store your prompts to help improve GPT Engineer? (y/n)"
)
while answer.lower() not in ("y", "n"):
answer = input("Invalid input. Please enter y or n: ")
if answer.lower() == "y":
path = Path(".gpte_consent")
path.write_text("true")
print(colored("Thank you️", "light_green"))
print()
print(
"(If you no longer wish to participate in data collection, delete the file .gpte_consent)"
)
return True
else:
print(
colored(
"No worries! GPT Engineer will not collect your prompts. ❤️",
"light_green",
)
)
return False
[docs]
def get_session() -> str:
"""
Retrieves or generates a unique identifier for the current user session.
This function attempts to read a unique user ID from a temporary file. If the file does not exist, it generates a new random ID, writes it to the file, and returns it. This ID is used to uniquely identify the user's session.
Returns
-------
str
A unique identifier for the user session.
"""
path = Path(tempfile.gettempdir()) / "gpt_engineer_user_id.txt"
try:
if path.exists():
user_id = path.read_text()
else:
# random uuid:
user_id = str(random.randint(0, 2**32))
path.write_text(user_id)
return user_id
except IOError:
return "ephemeral_" + str(random.randint(0, 2**32))