PyRFF - Random Fourier Features with Numpy

About - TLDR

• Use this repo if you're too lazy to extract features
• But still want good performance (in terms of speed and model "accuracy")
• The only dependency is NumPy.
• Extremely Lightweight
• Use this for Millions of predictions/day.

This repository extends this repo to sequential features.

Usage

from PyRFF import get_features_sequential, get_features
import numpy as np

# List of variable size vectors
sequential = [np.random.normal(size=(np.random.randint(1, 12), 4))
              for i in range(4)]
# Get Sequential Features
feat = get_features_sequential(
    sequential,  # Input List
    123,  # Random Seed
    # Feature Type orf (Orthogonal Random Feature) or rff (Random Fourier Features)
    "orf",
    6,  # Output Feature Size // 2
    0.1,  # Standard Deviation for Random Kernel
    max_length=6  # Maximum padded size for input vector (time dimension)
)

print(feat.shape)  # (4, 12)

Installation

pip install py-rff

• Check "Example.py" for the starter code.

Authors

This Repo

Surya Kant Sahu - surya.@gmail.com

Original Author

Tetsuya Ishikawa - tiskw111@gmail.com