Adaptive Online Block Subsampling for Time Series Data (OTSS)

OTSS is an algorithm to subsample time series while retaining their statistical properties.

Documentation

• Subsampler
• Breakpoint Detection
• Competing Subsamplers
  • Empirical Bernstein
  • Uniform Subsampler

The pipeline works by simultaneously delimiting the piecewise stationary time series, and subsampling an optimal size subsample from the individual pieces. Multiprocessing is leveraged to ensure scalability.

Subsampler

There are 4 typers of subsamplers:

online_subsampler(data, queue, return_queue, index_queue,  block_size, epsilon, delta, variable_range, max_iteration = 1000)

Breakpoint Detection

breakpoint_detection(data, queue, threshold = 1, drift = 0, ending = False, show = True, ax = None)

Competing Subsamplers

Empirical Bernstein

empirical_bernstein(data, epsilon, delta, variable_range)

Uniform Subsampler

uniform_sampler(data, rate = 2)

Example

License

MIT