This package contains implementations of subspace estimation and tracking techniques, with a focus on those techniques that are compatible with missing data.

Currently implemented are:

• GROUSE
• PETRELS

Example Usage (fixed subspace)

import numpy as np

from subtracking import StaticSubspace, Grouse, Petrels, calc_subspace_proj_error


dg = StaticSubspace(50, 5)
tracker = Petrels(50, 5, .98 )

ob_count = 10000

for i in range(ob_count):
    sampling_vec = np.ones((50,1))
    # set 4 random entries to be missing
    for i in range(4):
        idx = np.random.randint(0, high=50)
        sampling_vec[idx,0] = 0

    observation = dg.next_observation()
    tracker.consume(observation, sampling_vec)

    error = calc_subspace_proj_error(dg.U, tracker.U)

    print(error)