Quick start

This fits a small GPTree on six example founder profiles, draws the tree and prints a prediction for each profile. Run it in Jupyter, which allows await at the top level, with OPENAI_API_KEY set and Graphviz installed.

import pandas as pd
from IPython.display import Image, display
from think_reason_learn.gptree import GPTree
from think_reason_learn.core.llms import OpenAIChoice

X = pd.DataFrame({"founder_info": [
    "Alex is a serial entrepreneur with two successful exits and expertise in AI.",
    "Jordan graduated top of class from Oxford but has no business experience.",
    "Taylor has 10 years in finance, raised a seed round quickly and built a strong team.",
    "Casey started a company out of high school and has faced several failures.",
    "Morgan is a former Google engineer with machine learning patents and VC backing.",
    "Seraphine has a first-class degree in Minerals Engineering and a strong mining network.",
]})
y = ["successful", "failed", "successful", "failed", "successful", "successful"]

llm = [OpenAIChoice(model="gpt-4o-mini")]
tree = GPTree(qgen_llmc=llm, critic_llmc=llm, qgen_instr_llmc=llm, max_depth=2)
await tree.set_tasks(task_description="Predict whether a founder succeeds or fails from their background.")

async for node in tree.fit(X, y, reset=True):
    pass  # each step yields the node just built

display(Image(tree.view_node(tree.get_root_id())))
print(tree.get_questions())

async for index, question, answer, node_id, usage in tree.predict(X):
    print(index, question, answer)

Other providers work the same way. Use AnthropicChoice, GoogleChoice or XAIChoice from think_reason_learn.core.llms with the matching API key. The API reference covers Random Rule Forest and the other methods.