bp_osd: Belief propagation with ordered statistics decoding post-processing

[!WARNING]
This repo is no longer mantained. The functionality has been moved to https://github.com/MarcSerraPeralta/dem-decoders.

This is a wrapper for the ldpc.bposd_decoder to work with stim.

Example of usage:

import stim
import numpy as np

from pymatching import Matching
from bp_osd import BP_OSD

circuit = stim.Circuit.generated(code_task="surface_code:rotated_memory_z", 
                                 distance=3, 
                                 rounds=5, 
                                 after_clifford_depolarization=0.01)

sampler = circuit.compile_detector_sampler()
defects, log_flips = sampler.sample(shots=1_000_000, separate_observables=True)

mwpm = Matching(circuit.detector_error_model(decompose_errors=True))
predictions = mwpm.decode_batch(defects)
print(np.average(predictions != log_flips)) 
# logical error probability ~ 3.45%

bp_osd_cs = BP_OSD(circuit.detector_error_model(), osd_method="osd_cs", osd_order=40)
predictions = bp_osd_cs.decode_batch(defects)
print(np.average(predictions != log_flips)) 
# logical error probability ~ 3.24%