AutoML prediction engine — give it data, it finds the best model. Wins 57% on 513 OpenML datasets vs AutoGluon and FLAML.
Project description
🎯 Orcetra — Automated Prediction Engine
AutoML that beats FLAML on 78.4% of datasets. Verified on 513 OpenML benchmarks with equal 30-second compute budget.
Live Dashboard · Website · Paper (coming soon)
What Is Orcetra?
Orcetra is an automated prediction engine that combines intelligent model search with meta-learning to outperform established AutoML frameworks. It works on any tabular dataset — no manual feature engineering required.
Benchmark Results
Orcetra vs FLAML (513 OpenML datasets, strict 30s budget each)
| Orcetra Wins | FLAML Wins | Tie | |
|---|---|---|---|
| Overall | 402 (78.4%) | 75 (14.6%) | 36 (7.0%) |
| Classification (382) | 299 (78.3%) | 49 (12.8%) | 34 (8.9%) |
| Regression (131) | 103 (78.6%) | 26 (19.8%) | 2 (1.5%) |
By Dataset Size
| Scale | Orcetra | FLAML | Tie |
|---|---|---|---|
| Small (<5K samples) | 73 (86%) | 7 (8%) | 5 |
| Medium (5-50K) | 260 (79%) | 48 (15%) | 21 |
| Large (>50K) | 69 (70%) | 20 (20%) | 10 |
Compute Fairness
Both systems receive exactly 30 seconds. Orcetra's baseline evaluation phase counts toward the budget.
| Median Time | |
|---|---|
| Orcetra | 30.0s |
| FLAML | 31.3s |
Polymarket Prediction Benchmark
| Category | Beat Rate | Predictions |
|---|---|---|
| Overall | 66.7% | 2,932 |
| Sports | 81% | — |
| Politics | 100% | — |
| Economy | 100% | — |
Live tracking of 22,000+ active predictions across 8 categories.
How It Works
Phase 1: Baseline Pool → Evaluate RF, GBC, XGB, LightGBM, LogReg, etc.
Phase 2: AutoResearch → LLM-guided or random search for better configs
Phase 3: Meta-Learning → Strategy knowledge base from prior dataset wins
Phase 4: Validation → Cross-validated final evaluation
Key insight: Instead of just searching hyperparameters (like FLAML/Auto-sklearn), Orcetra searches across model families, preprocessing strategies, and feature transformations simultaneously.
Quick Start
git clone https://github.com/orcetra/orcetra.git
cd orcetra
pip install -r requirements.txt
Run on any dataset
from orcetra.core.agent import RandomSearchAgent
from orcetra.models.registry import get_baselines
from orcetra.metrics.base import get_metric
# Your data
metric = get_metric("accuracy") # or "mse" for regression
baselines = get_baselines("classification")
# Phase 1: Baselines
for name, fn in baselines.items():
score = fn(data_info, metric)
# Phase 2: AutoResearch loop
agent = RandomSearchAgent(task_type="classification")
proposal = agent.propose(data_info, metric, best_score, best_model, iteration)
Run Polymarket predictions
python batch_tracker.py predict # Generate predictions
python auto_check.py # Verify against outcomes
python scripts/gen_dashboard.py # Update dashboard
Architecture
src/orcetra/
├── core/ → Search agents (Random, LLM-guided)
├── models/ → Model registry & baselines
├── metrics/ → Evaluation metrics (accuracy, MSE, Brier)
└── meta/ → Strategy knowledge base
experiments/
├── openml_benchmark.py → Full OpenML benchmark suite
├── flaml_pilot_v2.py → Head-to-head vs FLAML
└── flaml_strict_30s.py → Fair compute-budget comparison
batch_tracker.py → Polymarket prediction pipeline
auto_check.py → Outcome verification
live_tracker.py → LLM-powered deep analysis
Roadmap
- Baseline model pool (RF, GBC, XGB, LightGBM, HistGBM, LogReg, etc.)
- Random search agent
- OpenML benchmark framework (513 datasets, 78.4% win rate vs FLAML)
- FLAML head-to-head comparison (strict 30s, compute-fair)
- Polymarket live prediction pipeline (22K+ markets)
- LLM-guided search agent (Groq/OpenAI)
- Meta-learning knowledge base
- MLE-bench evaluation
- Automated feature engineering module
- Multi-framework comparison (Auto-sklearn, H2O, AutoGluon)
Team
| Name | Role | Affiliation |
|---|---|---|
| Beibei Li | Chief Scientist | Carnegie Mellon University |
| Xiyang Hu | Researcher | Carnegie Mellon University |
| Luyi Ma | Researcher | Carnegie Mellon University |
| Kai Zhao | Researcher | New York University |
| Guilin Zhang | Researcher | George Washington University |
License
MIT License
"The best model isn't the one you know — it's the one you discover."
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file orcetra-0.1.0.tar.gz.
File metadata
- Download URL: orcetra-0.1.0.tar.gz
- Upload date:
- Size: 113.1 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.12.3
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
dcd712c7e3263ad0cc5359386270bcbcdc4139a626dd9b9d4dbf04f5eb06988d
|
|
| MD5 |
181a5f91904a3638cde146cf4fdf2f7d
|
|
| BLAKE2b-256 |
94127ac2904aa9fff1d2a3a0e234a8c4d95bd1d7dbe3afb6829319b435b4afa9
|
File details
Details for the file orcetra-0.1.0-py3-none-any.whl.
File metadata
- Download URL: orcetra-0.1.0-py3-none-any.whl
- Upload date:
- Size: 81.0 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.12.3
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
149d2048e55da4fdbb5d39cefac6f435c3170c0f1f42ba2a63269b0bed97c435
|
|
| MD5 |
0def91ac5770183ea9208fcc7dc7b74a
|
|
| BLAKE2b-256 |
e4efc279045d7aeb3d9d0c35b80584ff64b11b4398fbff7d18ef7978f3b51193
|