Metadata-Version: 2.4
Name: Orange3-TabH2O
Version: 0.1.0
Summary: Orange Data Mining add-on for H2O.ai's TabH2O foundation model.
Home-page: https://github.com/youraccount/orange3-tabh2o
Author: Carlos
License: BSD-3-Clause
Keywords: orange3 add-on,tabh2o,tabpfn,h2o,foundation model
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Education
Classifier: Intended Audience :: Science/Research
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Description-Content-Type: text/markdown
Requires-Dist: Orange3>=3.34
Requires-Dist: requests>=2.25
Requires-Dist: numpy
Dynamic: author
Dynamic: classifier
Dynamic: description
Dynamic: description-content-type
Dynamic: home-page
Dynamic: keywords
Dynamic: license
Dynamic: requires-dist
Dynamic: summary

# Orange3-TabH2O

An Orange Data Mining add-on that calls the [TabH2O](https://tabh2o.h2oai.com) foundation-model API to predict on new tabular observations.

Adds one widget — **TabH2O Predict** — to a new **TabH2O** category in the Orange toolbox.

## What it does

- Takes a training table (with a target column) and a test table (rows you want predicted)
- Sends both to `https://tabh2o.h2oai.com/api/v1/predict`
- Returns the test table with predictions appended as meta columns
  - **Classification:** predicted class + per-class probabilities
  - **Regression:** predicted value + confidence-interval bounds
  - **Time-series regression:** same, with a `time_column` you select

The widget runs the API call in a background thread (so Orange doesn't freeze) and handles `429 Rate Limited` automatically with one polite retry honoring `Retry-After`.

## Install (development mode)

You need Orange 3.34 or newer. From this directory:

```bash
pip install -e .
```

Then launch Orange — a new **TabH2O** category will appear in the toolbox on the left.

## Use

1. Get a free API key at https://tabh2o.h2oai.com (sign in with LinkedIn or Google).
2. Drop a **TabH2O Predict** widget on the canvas.
3. Paste your API key into the widget.
4. Connect a `File` (or `Datasets`, etc.) widget carrying your **training data** to its first input (`Data`).
5. Connect another `File` widget carrying the **rows you want predicted** to the second input (`Test Data`). The test table should have all the same columns as the training table *except* the target.
6. Pick the target column. If your training table already has a class set in Orange, it's used automatically.
7. Hit *Apply*. The output is your test table with prediction columns appended — feed it to `Data Table` to inspect.

## Free-tier limits

- 2 requests/minute, 20/day, 500/month
- ≤100,000 rows and ≤100 columns per request

For classroom use with more than ~10 students predicting at the same time you'll want to ask H2O.ai about a higher tier.

## API key storage

By default the API key is stored in your **per-user Orange settings** and is *not* written into the workflow `.ows` file. If you tick "Remember in workflow", the key is saved into the workflow and travels with it — convenient for a single-user demo, dangerous if you share the file. The widget shows a warning when this is on.

## License

BSD-3-Clause.
