Frypto is a Python package designed to compute a set of financial time series features, including price-based, volume-based, volatility, momentum, trend, and statistical features.
- In the G-Research competition, the top 3 winners demonstrated the critical role feature engineering plays in predicting real-world crypto market data.
- Frypto saves over 100 hours of manual feature creation and uses memory optimization tools to handle large datasets smoothly.
pip install fryptoLoad your cryptocurrency data (or download data with yfinance as shown below).
from frypto import AllFeatures
import yfinance as yf
# Download BTC-USD data from Yahoo Finance (or use your dataset)
data = yf.download('BTC-USD', period='max', interval='1d')
# Initialize Frypto and compute features
allfeatures = AllFeatures(data)
features = allfeatures.compute() # Returns a pd.DataFrame with the new features
# View the top rows of the generated features
print(features.head()) # New DataFrame with 41 featuresFrypto computes a range of financial time series features, including price changes, spreads, rolling statistics, momentum indicators, trend indicators, volatility measures, volume-based features, and lagged features.
- Price and Spread Features:
Price_change,next_log_return,high_low_spread,close_open_spread - Volatility Features:
rolling_std,upper_band,lower_band,ATR - Momentum Indicators:
RSI,MACD,Signal_Line,SMA,EMA,ROC - Statistical Features:
Skew_20,Kurtosis_20,ZScore_20 - Volume-Based Features:
volume_change,OBV - Lagged and Rolling Features:
lag_1,lag_2,lag_3,rolling_mean,rolling_max,rolling_min(for windows 5, 10, and 20) - Trend Indicators:
+DI,-DI,ADX,support_line,resistance_line,tenkan_sen,kijun_sen,senkou_span_a,senkou_span_b,chikou_span
import numpy as np
import pandas as pd
import yfinance as yf
from price import PriceFeatures
from volatility import VolatilityFeatures
from momentum import MomentumFeatures
from statistical_features import StatisticalFeatures
from volume import VolumeFeatures
from trend_features import TrendFeatures
from lag_rolling import LagRollingFeatures
# Download historical BTC-USD data
data = yf.download('BTC-USD', period='max', interval='1d')
close = data['Close'].values
high = data['High'].values
low = data['Low'].values
volume = data['Volume'].values
open_ = data['Open'].values
# Initialize and compute Price Features
price_features = PriceFeatures(close=close, high=high, low=low, open=open_)
price_df = price_features.compute()
print("Price Features:\n", price_df.head())
# Initialize and compute Volatility Features with a specified window
volatility_features = VolatilityFeatures(close=close, high=high, low=low)
volatility_df = volatility_features.compute(window=15)
print("Volatility Features:\n", volatility_df.head())
# Initialize and compute Momentum Features, with specified windows for RSI and MACD
momentum_features = MomentumFeatures(close=close)
momentum_df = momentum_features.compute(window=15, rsi_window=14, macd_windows=(12, 26, 9))
print("Momentum Features:\n", momentum_df.head())
# Initialize and compute Statistical Features with a specified rolling window
statistical_features = StatisticalFeatures(close=close)
statistical_df = statistical_features.compute(window=20)
print("Statistical Features:\n", statistical_df.head())
# Initialize and compute Volume Features
volume_features = VolumeFeatures(close=close, volume=volume)
volume_df = volume_features.compute()
print("Volume Features:\n", volume_df.head())
# Initialize and compute Trend Indicators with a specified window
trend_features = TrendFeatures(close=close, high=high, low=low)
trend_df = trend_features.compute(window=15)
print("Trend Features:\n", trend_df.head())
# Initialize and compute Lagged and Rolling Statistics with specified lags and windows
lag_rolling_features = LagRollingFeatures(close=close)
lag_rolling_df = lag_rolling_features.compute(lags=[1, 2, 3], windows=[5, 10, 20])
print("Lag and Rolling Features:\n", lag_rolling_df.head())
# Combine all computed features into a single DataFrame
all_features_df = pd.concat([
price_df,
volatility_df,
momentum_df,
statistical_df,
volume_df,
trend_df,
lag_rolling_df
], axis=1)
print("All Computed Features:\n", all_features_df.head())Frypto significantly enhances model performance by generating highly informative features. Below is a comparison of the top 20 feature importances for models trained with and without Frypto features:
Explore the example notebooks to learn more about Frypto:
- Getting Started with Frypto: Step-by-step guide to setting up and using Frypto.
- Exploring Individual Feature Types: Detailed exploration of each feature type and its applications. kaggle kernel | Crypto Feature Engineering: 29M Rows in 8 Minutes: Showcase of Frypto's performance and scalability on a massive dataset, demonstrating the generation of 42 features in under 9 minutes.
git clone https://github.com/NotAbdelrahmanelsayed/frypto
cd fryptoEnsure virtualenv is installed:
pip install virtualenvThen, create and activate a virtual environment:
Linux
virtualenv venv
source venv/bin/activateWindows Powershell
virtualenv venv
.\venv\Scripts\Activate.ps1pip install -r requirements.txtTo run unit tests:
pytestTo test multiple Python versions (3.7, 3.8, 3.9, 3.10, 3.11, 3.12):
toxIf you'd like to contribute, please fork the repository and open a pull request to the main branch.
This project is licensed under the MIT License.