trendpy (alpha)

Bayesian trend filtering micro library

I am currently redesigning the API and will release an alpha version in late August.

Documentation

The documentation can be found here.

Contributing

Contribution will be welcomed once a first stable release is ready. Contact Me

Import data

Data is imported from a file (trendpy only supports csv for now).

# import data from csv file (with dates and price) -- for now trendpy only
# support 1D time series

from trendpy.timeseries import Series

filename='data.csv'
fund=Series.from_csv(filename)

# plots time series
fund.plot()

Trend filtering

There is 1 trend filter available

• L1 filter (I plan to add a L2 filter in future releases)

Custom filters with more options to come in first stable release. The main advantage is that Bayesian trend filtering displays higher predictive power than frequentist methods (experiments to be added as example).

# trend filter with selected filter

fund.filter(number_simulations=30, burns=10)
fund.plot()

Requirements

These requirements reflect the testing environment. It is possible that trendpy will work with older versions.

• Python (3+)
• NumPy (1.12+)
• SciPy (0.13+)
• Pandas (0.19+)
• matplotlib (2.0+)
• statsmodels (0.6+)

Sources

Research papers that helped develop this library

• Locally adaptative regression splines (1997) - Mammen, van der Geer
• Asymptotic equivalence of non-parametric regression and white noise (1996) - Brown, Lo
• Postwar US business cycles: an empirical investigation (1997) - Hodrick Prescott
• Regression Shrinkage and Selection via the Lasso - (1996) Tibshirani
• Lasso Regression: Estimation and Shrinkage via Limit of Gibbs Sampling - (2015) Rayaratnam et al.
• Assessing Convergence of the Markov Chain Monte Carlo Method in Multivariate Case - (2005) Nogueira et al.

Support or Contact

Having trouble with trendpy? Check out our documentation.