asta-autodiscovery 0.2.2

Open-ended Scientific Discovery via Bayesian Surprise

Open-ended Scientific Discovery via Bayesian Surprise

Asta Autodiscovery is an autonomous agent that performs data exploration on arbitrary datasets. The agent will generate hypotheses and run experiments to test each one. Surprising outcomes generate follow-up hypotheses in a recursive exploration.

Link to our NeurIPS 2025 paper: AutoDiscovery: Open-ended Scientific Discovery via Bayesian Surprise

Installation

Requires Python 3.13 or newer.

pip install asta-autodiscovery

This installs the auto-discovery command-line tool.

Quick start

Point auto-discovery at one or more dataset files and describe what you want explored:

auto-discovery \
    --name "Plant growth study" \
    --description "Field trial measurements of plant height under varying fertilizer dosage" \
    --intent "Focus on dose-response relationships" \
    --n_experiments 20 \
    --out_dir ./results \
    data/measurements.csv data/treatments.csv

CSV/TSV column headers are detected automatically. Datasets can also be directories — every file under them will be included.

Dataset files/directories can have different descriptions for each one listed. Use a repeated --dataset_description parameter in place of the overall --description.

When the run finishes, a static HTML report is written to <out_dir>/report.

Common options

Flag	Description
--n_experiments	Number of experiments to run (required).
--out_dir	Output directory for results and the HTML report (required).
--name	Short title for the run.
--description	Context about the dataset: provenance, collection method, known gaps.
--domain	Research domain (e.g. Genomics).
--intent	High-level exploration guidance for the agent.
--dataset_description	Per-dataset description; repeat once per dataset, in order.
--exploration_weight	Higher = broader exploration (default 2.0).
--surprisal_width	Surprise threshold; lower = more sensitive (default 0.2).

Run auto-discovery --help to see the full set of options.

Authentication

The agent talks to model providers through their OpenAI-compatible endpoints. The provider is chosen per-model from the model name: anything starting with gemini is routed through Google Vertex AI; everything else (e.g. gpt-4o) goes to OpenAI. You only need to configure the providers for the models you actually select via --model, --belief_model, and --vision_model.

Gemini (Vertex AI)

Used when any of --model, --belief_model, or --vision_model is a gemini-* name. The defaults are Gemini models, so this is required unless you override all three.

Pick one of the following. In all cases, set the project (and optionally location) so the agent knows which Vertex endpoint to call:

export VERTEX_PROJECT_ID=your-gcp-project-id
export VERTEX_LOCATION=global   # optional; defaults to "global"

Service account key file (recommended for non-interactive use):

Create a service account in your GCP project, grant it the Vertex AI User role, download a JSON key, and point Google's standard ADC env var at it:

export GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account-key.json

User credentials via gcloud (recommended for local development):

gcloud auth application-default login

Static access token (short-lived, e.g. in CI): If present, this will take priority over the GOOGLE_APPLICATION_CREDENTIALS setting.

export VERTEX_ACCESS_TOKEN=$(gcloud auth print-access-token)

To bypass project/location lookup entirely, set VERTEX_OPENAI_BASE_URL to a fully-formed Vertex OpenAI-compatible endpoint URL.

OpenAI

Used when any of --model, --belief_model, or --vision_model is a non-Gemini name (e.g. gpt-4o, gpt-4o-mini).

export OPENAI_API_KEY=sk-...

Selecting models

Flag	What it controls	Provider
--model	Primary reasoning model used for hypothesis generation and analysis.	Gemini if name starts with gemini, else OpenAI.
--belief_model	Model used for belief updates over experimental outcomes.	Same routing.
--vision_model	Model used to interpret plots and figures emitted by experiments.	Same routing.

Mixing providers is supported — for example, --model gpt-4o --belief_model gemini-3-flash-preview will use OpenAI for the main loop and Vertex AI for belief updates, and both OPENAI_API_KEY and the Vertex variables must be set.

Citation

If you find this work useful, please cite:

@inproceedings{
agarwal2025autodiscovery,
title={AutoDiscovery: Open-ended Scientific Discovery via Bayesian Surprise},
author={Dhruv Agarwal and Bodhisattwa Prasad Majumder and Reece Adamson and Megha Chakravorty and Satvika Reddy Gavireddy and Aditya Parashar and Harshit Surana and Bhavana Dalvi Mishra and Andrew McCallum and Ashish Sabharwal and Peter Clark},
booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
year={2025},
url={https://openreview.net/forum?id=kJqTkj2HhF}
}