edfwriter not copying arrays perfectly. Random elements, most do not differ, differ from the numpy array used to create edf file by exactly -0.0984668518066405 or 0.0984668518066405 every time.

[NathanTBurgess]

edfwriter not copying arrays perfectly. Random elements, most do not differ, differ from the numpy array used to create edf file by exactly -0.0984668518066405 or 0.0984668518066405 every time.

Here is the most relevant function. Just know when I find the differences between the returned numpy array of the written file and the original numpy array you see those random differences by that exact amount scattered in the indicies.

def write_edf_slice(shm_original_names, signal_headers, f, output_edf_file_path, start_time, end_time, shm_concat_names, start_pos, n_elements, total_elements, channel_number):
    # Create a new EDF file writer for the time slice
    output_edf_writer = pyedflib.EdfWriter(output_edf_file_path, len(signal_headers), file_type=pyedflib.FILETYPE_EDFPLUS)

    original_signal_data = []
    shm_original_list = []

    for i in range(len(shm_original_names)):
        shm_original_list.append(shared_memory.SharedMemory(name=shm_original_names[i]))
        original_signal_data.append(np.ndarray((total_elements), dtype=np.float64, buffer=shm_original_list[i].buf))

    concatenated_signal = []
    shm_concat_list = []

    for i in range(len(shm_concat_names)):
        shm_concat_list.append(shared_memory.SharedMemory(name=shm_concat_names[i]))
        concatenated_signal.append(np.ndarray((n_elements), dtype=np.float64, buffer=shm_concat_list[i].buf, offset=start_pos * np.dtype(np.float64).itemsize))

    # Copy signal headers to the new EDF file
    for i in range(channel_number):
        signal_header = signal_headers[i]
        output_edf_writer.setSignalHeader(i, signal_header)
        start_index = int(start_time * f[i])
        end_index = int(end_time * f[i])
        original_data_current_channel = original_signal_data[i][start_index:end_index]
        np.copyto(concatenated_signal[i], original_data_current_channel)

    output_edf_writer.writeSamples(concatenated_signal)

    output_edf_writer.close()

    # Close shared memory segment
    for shm in shm_original_list:
        shm.close()

    for shm in shm_concat_list:
        shm.close()

[skjerns]

can you post a minimal reproducible sample with code?

else it's difficult to find the bug

[TrianecT-Wouter]

You are using a np.float64 ( (2^64 steps) datatype. EDF stores float16 (2^16 steps) ranging from -digital_minimum to +digital_minimum and using physical_minumum and physical_maximum. This might explain (some) of the differences you see?