neuralk 1.3.0

Overview

The Neuralk SDK provides Python developers with a scikit-learn compatible interface to access Seldon, a foundation model specifically designed for tabular classification tasks.

Key Features:

• Zero hyperparameter tuning - Strong baseline performance out of the box
• Scikit-learn compatible - Works with pipelines, cross-validation, and familiar fit/predict interface
• Flexible deployment - Cloud API or on-premise server
• Mixed feature types - Pandas and polars DataFrames are sent as Parquet, so categorical and string columns keep their types

Installation

pip install neuralk

Requirements: Python 3.11+

Quick Start

1. Get your API key

neuralk login

This displays instructions to create your account at prediction.neuralk-ai.com/register.

2. Set your API key

export NEURALK_API_KEY=nk_live_your_api_key_here

3. Make your first prediction

from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from neuralk import SeldonClassifier
from neuralk.datasets import two_moons

# Load example dataset
X, y = two_moons()
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)

# Create classifier and predict
clf = SeldonClassifier()
clf.fit(X_train, y_train)
predictions = clf.predict(X_test)

print(f"Accuracy: {accuracy_score(y_test, predictions):.2%}")

Using with scikit-learn pipelines

from sklearn.pipeline import make_pipeline
from sklearn.impute import SimpleImputer
from skrub import TableVectorizer
from neuralk import SeldonClassifier

clf = make_pipeline(
    TableVectorizer(),
    SimpleImputer(),
    SeldonClassifier(),
)

clf.fit(X_train, y_train)
predictions = clf.predict(X_test)

On-premise deployment

from neuralk import SeldonClassifier

clf = SeldonClassifier(host="http://your-server:8000")
clf.fit(X_train, y_train)
predictions = clf.predict(X_test)

Explaining predictions

Understand why the model predicts what it does, using Shapley values and interactions (powered by shapiq). Install the optional extra (requires Python 3.12+):

pip install "neuralk[explain]"

from neuralk.explainability import SeldonExplainer

clf = SeldonClassifier().fit(X_train, y_train)
explainer = SeldonExplainer(clf, background=X_train)

# Local: explain a single prediction
local = explainer.explain(X_test[0])
explainer.plot_force(local)

# Global: aggregate feature importance over many predictions
global_ = explainer.explain_global(X_test[:50])
print(global_.feature_importances(normalize=True))
global_.plot_bar()

Note: Seldon predicts via remote API calls that consume credits, and explainers evaluate the model many times. Keep background small and the estimation budget modest while iterating.

Available Models

Documentation

• Getting Started Guide
• Understanding In-Context Learning
• Advanced Configuration
• Context Selection Strategies
• Example Gallery

Citation

If you use Neuralk in your research, please cite:

@software{neuralk2026sdk,
    title = {Neuralk: A Foundation Model for Industrial Tabular Data},
    author = {Neuralk AI},
    year = {2026},
    url = {https://www.neuralk-ai.com/}
}

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Contact

• Questions & Feature Requests: alex@neuralk-ai.com
• Collaborations: antoine@neuralk-ai.com