Using AI to "type" (really validate) Python code. Tries to be "Ambient AI". Allows you to do things like:
isinstance("Julia", PersonName)
>>> True
isinstance("Idaho", PersonName)
>>> FalseThe typing module in Python helps the user understand the codebase better.
They are useful for distinguishing between ints and strings, for example.
The aityping module helps do "fuzzy typing".
This is really run-time validation.
However, it's bundled up as a type so that you can use it with Pydantic Models easily.
pip install aityping
First, you need to set your OpenAI key:
import os
os.environ["OPENAI_API_KEY"] = 'YOUR_OPENAI_API_KEY'Next we import the get_ai_type function:
from aityping import get_ai_typeAfter that is done we can create a "type" with some fuzzy logic
PersonName = get_ai_type("is a persons name")We can now use this to do type checking as we would with any other type.
isinstance("Julia", PersonName)
>>> True
isinstance("Idaho", PersonName)
>>> FalseWe can also use it to type a Pydantic model and then have that do validation.
from pydantic import BaseModel
class Record(BaseModel):
name: PersonName
Record(name='Julia')
>>> Record(name='Julia')
Record(name='Idaho')
>>> ValidationError: 1 validation error for Record
>>> name
>>> Flagged as invalid. (type=type_error)By default, this will use the following LLM configuration:
ChatOpenAI(model="gpt-3.5-turbo", temperature=0)If you want to use a different configuration, you easily can!
from langchain.chat_models import ChatOpenAI
PersonName = get_ai_type("is a persons name", llm=ChatOpenAI(model="gpt-4"))