⚛️ Qkabrine AutoML

An open-source framework that unifies intelligent architecture search, quantum kernel methods, and variational circuits in a single AutoML pipeline.

Created by Eric Jagwara at Solid Elf Labs

pip install qkabrine-automl

from qkabrine_automl import QkabrineAutoML

automl = QkabrineAutoML(task='classification', search_strategy='bayesian')
automl.fit(X_train, y_train)
automl.leaderboard()
preds = automl.predict(X_test)
print(automl.export_qasm())

What Makes This Different

Existing quantum ML tools force you to choose a circuit, choose an encoding, choose a training strategy, and hope it works. Qkabrine AutoML searches across all of these simultaneously:

Key Features

1. Five search strategies in one API — grid, random, Bayesian (GP + Expected Improvement), evolutionary (genetic algorithm), and successive halving (HyperBand-inspired).

2. Joint search over architecture × encoding × hyperparameters — most QAS tools only search over gate arrangements. We co-optimize the complete pipeline.

3. Quantum kernels as first-class citizens — the search considers both variational circuits and quantum kernel + SVM methods, comparing them on equal footing.

4. Circuit surgery — post-search pruning of near-identity rotation gates and simplification of redundant gate pairs, reducing circuit depth for NISQ deployment.

5. True multi-class classification — qml.probs() with cross-entropy loss trains all classes simultaneously.

6. Expressibility & entangling capability metrics — KL-divergence expressibility and Meyer-Wallach entanglement for circuit analysis.

7. Training dynamics — DQFIM trainability prediction, barren plateau monitoring, and quantum natural gradient optimization.

8. Noise-aware training for NISQ hardware readiness + QASM export for deployment.

Quickstart

Binary Classification

from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
from qkabrine_automl import QkabrineAutoML

X, y = load_breast_cancer(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

automl = QkabrineAutoML(
    task='classification',
    n_qubits=4,
    max_layers=2,
    search_strategy='bayesian',
    encodings=('angle', 'iqp'),
    feature_reduction='pca',
)
automl.fit(X_train, y_train)
automl.leaderboard()
print(f"Test accuracy: {automl.score(X_test, y_test):.4f}")

Multi-Class Classification

from sklearn.datasets import load_iris
from qkabrine_automl import QkabrineAutoML

X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

automl = QkabrineAutoML(
    task='classification',
    n_qubits=4,
    search_strategy='evolutionary',
)
automl.fit(X_train, y_train)

Regression

automl = QkabrineAutoML(task='regression', search_strategy='bayesian')
automl.fit(X_train, y_train)

Save and Load

automl.save('my_model.pkl')
loaded = QkabrineAutoML.load('my_model.pkl')
loaded.predict(X_test)

Search Strategies

Circuit Architectures (12)

Training Dynamics

from qkabrine_automl import DataQuantumFisherMetric, BarrenPlateauMonitor

# Predict trainability before training
dqfim = DataQuantumFisherMetric(n_qubits=4)
metrics = dqfim.predict_generalization(circuit_fn, X, n_params=12)
print(f"Trainability: {metrics.trainability_score:.3f}")

# Monitor for barren plateaus during training
monitor = BarrenPlateauMonitor(threshold=1e-7, auto_surgery=True)

API Reference

QkabrineAutoML

Methods

Links

• Homepage: qkabrine.online
• Author: Eric Jagwara
• Lab: Solid Elf Labs
• Repository: github.com/ericjagwara/qkabrine
• Issues: github.com/ericjagwara/qkabrine/issues

License

MIT — Copyright (c) 2026 Eric Jagwara, Solid Elf Labs