Discrepancy in BpOsd decoder default_method between Python bindings docs and underlying C++ implementation

[jfoliveiraramos]

closes #99

Description

There appears to be a mismatch between the Python bindings documentation and the actual behaviour in the underlying C++ code for the BpOsd decoder.

In the Python stub (src_python/ldpc/bp_decoder/__init__.pyi, line 408), the __cinit__ method is depicted as:

def __cinit__(self, pcm: Union[np.ndarray, spmatrix], error_rate: Optional[float] = None,
             error_channel: Optional[List[float]] = None, max_iter: Optional[int] = 0, bp_method: Optional[str] = 'minimum_sum',
             ms_scaling_factor: Optional[float] = 1.0, cutoff: Optional[float] = np.inf, sigma: float = 2.0, **kwargs): ...

However, it seems that the actual C++ implementation is still using product_sum as the default (src_cpp/bp.hpp, line 77)

BpDecoder(
        BpSparse &parity_check_matrix,
        std::vector<double> channel_probabilities,
        int maximum_iterations = 0,
        BpMethod bp_method = PRODUCT_SUM,
        ...

Changes

Assuming the documentation reflects the intended use, these changes change the default value for bp_method for the BPDecoder class in src_cpp/bp.hpp.

[quantumgizmos]

Hi Joao,

Thanks for pointing this out. I'll test later today.

Joschka