• What is IBL
• Examples
• How to Use
• Inductive-bias Learning Models
• Supported LLMs
• Contributor
• Backstory

IBL (Inductive-bias Learning) is a new machine learning modeling method that uses LLM to infer the structure of the model itself from the data set and outputs it as Python code. The learned model (code model) can be used as a machine learning model to predict a new dataset.In this repository, you can try different learning methods with IBL.(Currently only binary classification with simple methods is available.)

• Currently, only binary classification is supported.

Use the link below to try it out immediately on Google colab.

• Binary classification

  • IBL

    • OpenAI:
    • Claude:

  • IBLbagging

    • OpenAI:
    • Claude:

• Installation and Import

pip install iblm

import iblm

• Setting

  • OpenAI

    os.environ["OPENAI_API_KEY"] = "YOUR_API_KEY"
    
    ibl = iblm.IBLModel(api_type="openai", model_name="gpt-4-0125-preview", objective="binary")

  • Azure OpenAI

    os.environ["AZURE_OPENAI_KEY"] = "YOUR_API_KEY"
    os.environ["AZURE_OPENAI_ENDPOINT"] = "xxx"
    os.environ["OPENAI_API_VERSION"] = "xxx"
    
    ibl = iblm.IBLModel(api_type="azure", model_name="gpt-4-0125-preview", objective="binary")

  • Google API

    os.environ["GOOGLE_API_KEY"] = "YOUR_API_KEY"
    ibl = iblm.IBLModel(api_type="gemini", model_name="gemini-pro", objective="binary")

  • Anthropic API

    os.environ["ANTHROPIC_API_KEY"] = "YOUR_API_KEY"
    ibl = iblm.IBLModel(api_type="", model_name="", objective="binary")

• Model Learning
  Currently, only small amounts of data can be executed.

  code_model = ibl.fit(x_train, y_train)
  
  print(code_model)

• Model Predictions

  y_proba = ibl.predict(x_test)

• Inductive-bias Learning
  Normal Inductive-bias Learning

  from iblm import IBLBaggingModel
  
  iblm = IBLModel(
      api_type="openai",
      model_name="gpt-4-0125-preview",
      objective="binary"
      )

• Inductive-bias Learning bagging
  Sampling data from a given dataset, we create multiple models, and the average of these models is used as the predicted value.

  from iblm import IBLBaggingModel
  
  iblbagging = IBLBaggingModel(
      api_type="openai",
      model_name="gpt-4-0125-preview",
      objective="binary",
      num_model=20,  # Number of models to create
      max_sample = 2000,  # Maximum number of samples from the data set
      min_sample = 300,　　# Minimum number of samples from the data set
      )

• OpenAI
  • gpt-4-0125-preview
  • gpt-3.5-turbo-0125
• Azure OpenAI
  • gpt-4-0125-preview
  • gpt-3.5-turbo-0125
• Google
  • gemini-pro
• Anthropic
  • claude-3-opus-20240229
  • claude-3-sonnet-20240229

• @t-ymbys
• @cn47

If you find this repo helpful, please cite the following papers:

@article{tanaka2023inductive,
  title={Inductive-bias Learning: Generating Code Models with Large Language Model},
  author={Tanaka, Toma and Emoto, Naofumi and Yumibayashi, Tsukasa},
  journal={arXiv preprint arXiv:2308.09890},
  year={2023}
}