Skip to content

README.md

Latest commit

 

History

History
98 lines (78 loc) · 3.01 KB

README.md

File metadata and controls

98 lines (78 loc) · 3.01 KB

DOI

⚡ CFIRE: Cross-Representation Feature Extraction for Time Series Classification

CFIRE (Cross-Representation Feature Extraction) is a feature-based time series classification framework that leverages a diverse set of time-domain, frequency-domain, and transformation-based representations. It systematically extracts, combines, and optimizes features across representations to achieve state-of-the-art classification performance.

🚀 Key Features

  • 🔁 Cross-representation feature extraction (Time, Frequency, Derivative, DWT, FFT, Hilbert, etc.)
  • 🧠 Support for popular feature sets: Catch22, TSFresh
  • 🔍 Integrated feature redundancy reduction and optimization
  • ⚙️ Built-in classifier selection and benchmarking (e.g., ExtraTrees, XGBoost, SVM)
  • 🧪 Parallelized feature extraction for scalable performance
  • 📊 Comprehensive experimental suite for reproducible research

📦 Installation

git clone https://github.com/alagoz/cfire.git
cd cfire
pip install -r requirements.txt

🚀 Quick Start Guide

1. Install Requirements

pip install numpy pandas scikit-learn aeon tsfresh PyWavelets dtaidistance

2. Load Any UCR Dataset via aeon

CFIRE supports any dataset from the UCR Time Series Archive via the aeon library.

To change the dataset, simply modify the name_dset in the script:

name_dset = "ArrowHead"  # Replace with any UCR dataset name

✅ Example Datasets
"ECG200"
"GunPoint"
"Coffee"
"Plane"
"ItalyPowerDemand"
"Chinatown"
"Beef"

To list all available datasets:

from aeon.datasets import get_dataset_names
print(get_dataset_names("classification"))

3. Run the Demo

python demo_.py

Sample Output on an Intel i7-11700 @2.50 GHz CPU with 8 cores and 32 GB RAM:

run:0 acc:0.9535, dur_tr_trans:4.59s, dur_clf_fit:0.15s, dur_te_trans:4.60s, dur_clf_pred:0.02s
run:1 acc:0.9302, dur_tr_trans:0.57s, dur_clf_fit:0.15s, dur_te_trans:0.42s, dur_clf_pred:0.00s
run:2 acc:0.9070, dur_tr_trans:0.17s, dur_clf_fit:0.11s, dur_te_trans:0.10s, dur_clf_pred:0.01s

🧠 How It Works

  • Load a dataset from the UCR archive using aeon.
  • Extract features using CFIRE from multiple domains (Fourier, Wavelet, Hilbert, etc.).
  • Train a classifier (e.g., Extremely Randomized Trees).
  • Evaluate accuracy and compute time.

📁 Project Structure

cfire/  
├── demo_.py       # Example: Run on any UCR dataset  
├── crossfire.py   # crossfire implementation (user-defined)  
├── README.md      # This file

⚙️ Customize Feature Extraction
You can control which feature categories are used in CFIRE:

model_feat = CFIRE(
    norms=True,
    stats=True,
    series=True,
    temp=True,
    multiprocessing=True
)

📬 Contact
For feedback or questions, open an issue or email celal.alagoz@gmail.com.