Skip to content

Latest commit

 

History

History
97 lines (69 loc) · 3.88 KB

README.md

File metadata and controls

97 lines (69 loc) · 3.88 KB
Raw

FEDOT.LLM

FEDOT.LLM is an LLM-based prototype for next-generation AutoML. It combines the power of Large Language Models with automated machine learning techniques to enhance data analysis and pipeline building processes.

Installation

  1. FEDOT.LLM is only available via github now. To install, clone the repository:

    git clone https://github.com/ITMO-NSS-team/FEDOT.LLM.git
cd FEDOT.LLM

  2. Install dependencies:

    pip install -r requirements.txt

How to Use

FEDOT.LLM provides a high-level API with simple interface through FedotAI class. It can be used to start the whole pipeline of LLM-powered dataset analysis and making predictions using FEDOT.

To use the API, follow these steps:

  1. Import FedotAI class

    from fedot_llm.main import FedotAI

  2. Initialize the FedotAI object. The required parameters are the following:

  • The dataset is a native fedot_llm.data.data.Dataset object that contains the dataset files. It can be initialized using specific loaders, such as the PathDatasetLoader.

  • The model is the chat model you want to use. You can use any chat model class from the langchain library. However, for the best experience, we recommend using models like gpt4o-mini or higher.

  • handlers is a list of output handlers to use. You can create your own output handler or use the pre-existing ones. For instance, JupyterOutput contains handlers for Jupyter notebooks. You can subscribe to all of them using JupyterOutput().subscribe.

To acquire predictions, use the ask method with a string description of the dataset and associated task in an arbitrary form.

# Import necessary modules and classes
from langchain_openai import ChatOpenAI
from pathlib import Path
from fedot_llm.data.loaders import PathDatasetLoader
from fedot_llm.main import FedotAI
from fedot_llm.output.jupyter import JupyterOutput

# Initialize the ChatOpenAI model
# Note: Make sure to set the OPENAI_TOKEN environment variable
llm = ChatOpenAI(model='gpt-4o-mini', base_url='https://models.inference.ai.azure.com', api_key=os.environ['OPENAI_TOKEN'])

# Set the path to the dataset
# Load the dataset using PathDatasetLoader
dataset_path = Path(module_path) / 'datasets' / 'Health_Insurance'
dataset = PathDatasetLoader.load(dataset_path)

# Define the task description for the model
msg="""Create a model that perform this task:
Our client is an insurance company that has provided health insurance to its customers.
They are interested in whether the policyholders (customers) from last year
will also be interested in the car insurance provided by the company."""

# Initialize FedotAI with the dataset, language model, and output handlers
fedot_ai = FedotAI(dataset=dataset, 
                model=llm,
                handlers=JupyterOutput().subscribe)

# Asynchronously process the task using FedotAI
# The loop continues until the task is completed
async for _ in fedot_ai.ask(message=msg):
    continue`

Examples and demo

You can use the example notebooks in the examples/by_datasets/ directory to get started. For instance, to run the Health_Insurance dataset example:

jupyter notebook examples/by_datasets/health_insurance.ipynb

You can also use the Streamlit web interface for a more interactive experience. To run it:

streamlit run streamlit-app.py

For more information on how to setup and run Streamlit app see STREAMLIT_README.md.

Development

If you want to contribute or set up a development environment, you can use the provided dev container.

This will set up a fully-featured development environment in a container, either in GitHub Codespaces or using VS Code's Dev Containers extension.

For more information see .devcontainer/README.md.