---
title: ROBERT
emoji: 🧪
colorFrom: indigo
colorTo: blue
sdk: gradio
sdk_version: 6.14.0
python_version: 3.12.12
app_file: app.py
pinned: false
license: mit
short_description: Automated and reproducible machine learning for chemistry
---

# ROBERT

Minimal web interface for **ROBERT** (Refiner and Optimizer of a Bunch of
Existing Regression Tools).

Upload a CSV dataset, select the target and name/ID columns, and run the full
ROBERT workflow. When the calculation finishes, the PDF report, figures and a
ZIP archive containing all results can be downloaded.

## Input

- A comma-, semicolon- or tab-separated CSV file.
- One target column (`y`).
- One column containing the datapoint names or IDs.
- Descriptor columns used to build the models.

The uploaded table is normalized to a comma-separated CSV before ROBERT is
started. Public deployments should only be used with non-confidential data.

## Links

- [ROBERT source code](https://github.com/jvalegre/robert/tree/ddg_branch)
- [Documentation](https://robert.readthedocs.io/)

ROBERT is distributed under the MIT License.

## Hardware

This version is adapted for **ZeroGPU** so that eligible free personal accounts
can host it without subscribing to PRO. The ROBERT calculation is exposed
through `@spaces.GPU(duration=300)` and must finish within five minutes.

ROBERT currently relies mainly on CPU-based scikit-learn workflows, so the
allocated GPU does not make the calculation faster. It is used here to satisfy
the ZeroGPU execution model. Free accounts currently receive five minutes of
ZeroGPU usage per day; unauthenticated visitors receive two minutes.