pydis-nn 1.0.1

A lightweight Python package for training neural networks to interpolate 5-dimensional numerical datasets

pydis_nn

A lightweight Python package for training and using neural networks to interpolate 5-dimensional numerical datasets. Built with TensorFlow/Keras and optimized for fast training and inference.

Features

• Data Handling: Load, validate, preprocess, and split 5D datasets (.pkl format)
• Neural Network: Configurable feedforward neural network with customizable architecture
• Training: Train models with optional validation data and early stopping
• Evaluation: Compute R² scores and MSE metrics on test sets
• Utilities: Generate synthetic 5D datasets for testing

Quick Start

Installation

pip install pydis_nn

Basic Usage

from pydis_nn import NeuralNetwork, load_and_preprocess

# Load and preprocess your dataset
data = load_and_preprocess('your_dataset.pkl', random_state=42)

# Create and train a neural network
model = NeuralNetwork(
    hidden_sizes=[64, 32, 16],  # 3 hidden layers
    learning_rate=0.001,
    max_iter=300,
    random_state=42
)

model.fit(
    data['X_train'],
    data['y_train'],
    X_val=data['X_val'],
    y_val=data['y_val']
)

# Make predictions
predictions = model.predict(data['X_test'])

# Evaluate performance
r2_score = model.score(data['X_test'], data['y_test'])
print(f"R² Score: {r2_score:.4f}")

Documentation

📚 Full Documentation: Read the Docs

The documentation includes:

• API Reference
• User Guides (Installation, Usage, Dataset Format)
• Performance Profiling
• Testing Information

Requirements

• Python >= 3.10
• NumPy >= 1.24.0
• TensorFlow >= 2.13.0
• scikit-learn >= 1.3.0
• SciPy >= 1.10.0

Dataset Format

Your dataset should be a .pkl (pickle) file containing a dictionary:

{
    'X': numpy.ndarray,  # Shape: (n_samples, 5) - exactly 5 features
    'y': numpy.ndarray   # Shape: (n_samples,) - target values
}

Features in Detail

Data Module (pydis_nn.data)

• load_dataset(): Load and validate 5D datasets from .pkl files
• split_data(): Split data into train/validation/test sets
• standardize_features(): Standardize features using training statistics
• load_and_preprocess(): Complete preprocessing pipeline

Neural Network Module (pydis_nn.neuralnetwork)

• NeuralNetwork: Configurable neural network class
  • Customizable hidden layer sizes
  • Adam optimizer with configurable learning rate
  • Early stopping support
  • Returns training history for visualization

Utilities (pydis_nn.utils)

• generate_sample_dataset(): Generate synthetic 5D datasets for testing

Performance

The package is optimized for fast training:

• 10,000 samples train in under 32 seconds on a MacBook Air M2
• Peak memory usage: ~5 MB during training
• Efficient sub-linear scaling with dataset size

See the Performance Profiling documentation for detailed benchmarks.

License

MIT License - See LICENSE file for details.

Author

Harvey Bermingham

Links

• Documentation: https://pydis-nn.readthedocs.io/
• Source Code: https://github.com/harveydgb/full-stack-nn (or your repository URL)