ccpfn 0.0.1

CCPFN: Causal Foundation Model for Continuous Treatments

Causal Foundation Models with Continuous Treatments

This is the repository for inference with CCPFN (Continuous Causal Prior-Fitted Network), a causal foundation model for use in domains with a continuous treatment variable. This is the inference repository; the full research reposistory (including training code and our prior) is forthcoming.

Quick Start

To install from source (a PyPI package is forthcoming), run the following:

git clone https://github.com/layer6ai-labs/CCPFN-inference.git
cd CCPFN-inference
pip install -e .

Example notebooks on CCPFN usage, including individual treatment-response curve reconstruction tasks, can be found in the notebooks directory.

Overview

CCPFN uses in-context learning (ICL) to estimate the effects of a continuously-varying treatment (for example, the dosage of a medication, or the sensitivity of economic outcomes to prices or rates). Specifically, it estimates the conditional expected potential outcome (CEPO), defined as $𝔼[Y(t) \mid X = x]$. Observational (historic) data is supplied as context, and queries are passed at inference time. No further training is required.