parsing.py

# -*- coding: utf-8 -*-
"""
This submodule contains a collection of helper to parse filenames and some fields in
*bb_binary* repositories to Python datastructures.

There are also some helper to convert some data representations like timeformats or ids.
"""
import os
from datetime import datetime
from bitarray import bitarray
import numpy as np
import iso8601
import pytz

_TIMEZONE = pytz.timezone('Europe/Berlin')


def dt_to_str(dt):
    dt = to_datetime(dt)
    isoformat = "%Y-%m-%dT%H:%M:%S"

    dt_str = dt.strftime(isoformat)
    if dt.microsecond != 0:
        dt_str += ".{:06d}".format(dt.microsecond)
    if dt.tzinfo is not None and dt.utcoffset().total_seconds() == 0:
        return dt_str + "Z"
    else:
        raise Exception("Got a datetime object not in UTC. Allways use UTC.")


def get_fname(camIdx, dt):
    dt = to_datetime(dt)
    return ("Cam_{cam}_{ts}").format(cam=camIdx, ts=dt_to_str(dt))


def get_video_fname(camIdx, begin, end):
    return get_fname(camIdx, begin) + "--" + dt_to_str(end)


def get_timezone():
    return _TIMEZONE


def int_id_to_binary(int_id, nb_bits=12):
    """Helper to convert an id represented as integer to a bit array.

    Warning:
        This function uses big-endian notation whereas :obj:`.DetectionDP` uses
        little-endian notation for `decodedId`.

    Arguments:
        int_id (int): the integer id to convert to a bit array

    Keyword Arguments:
        nb_bits (Optional int): number of bits in the bit array

    Returns:
        :obj:`np.array`: the bit array in big-endian notation
    """
    result = np.zeros(nb_bits, dtype=np.uint8)
    if int_id >= 2**nb_bits:
        raise Exception("Id {} overflows {} bits".format(int_id, nb_bits))
    pos = nb_bits - 1
    while pos >= 0:
        result[pos] = int_id & 1
        int_id >>= 1
        pos -= 1
    return result


def binary_id_to_int(decoded_id, threshold=0.5, endian='big'):
    """Helper to convert an id represented as bit array to an integer.

    Warning:
        :obj:`.DetectionDP` uses little-endian notation for `decodedId`.

    Note:
        This is a generic solution to the problem. If have to decode a lot of ids take a look on
        `numpy.packbits()
        <https://docs.scipy.org/doc/numpy/reference/generated/numpy.packbits.html>`_ and implement a
        vectorized version.

    Arguments:
        decoded_id (:obj:`list` of int or float): id as bit array

    Keyword Arguments:
        threshold (Optional float): `decoded_id` values >= threshold are interpreted as 1
        endian (Optional str): use either 'big' or 'little', default is 'big'

    Returns:
        int: the decoded id represented as integer
    """
    if endian == 'little':
        decoded_id = decoded_id[::-1]
    return int(bitarray([bit >= threshold for bit in decoded_id], endian=endian).to01(), 2)


def parse_cam_id(fname):
    fname = os.path.basename(fname)
    camIdx = fname.split("_")[1]
    return int(camIdx)


def parse_fname(fname):
    fname = os.path.basename(fname)
    try:
        return parse_video_fname(fname, 'iso')
    except:
        try:
            return parse_video_fname(fname, 'beesbook')
        except:
            camIdx, dtime = parse_image_fname(fname)
            return camIdx, dtime, dtime


def parse_image_fname_beesbook(fname):
    basename = os.path.basename(fname)
    name = basename.split('.')[0]
    name = name.replace('__', '_')
    _, camIdxStr, datetimeStr, usStr = name.split('_')

    camIdx = int(camIdxStr)
    year = int(datetimeStr[:4])
    month = int(datetimeStr[4:6])
    day = int(datetimeStr[6:8])

    hour = int(datetimeStr[8:10])
    minute = int(datetimeStr[10:12])
    second = int(datetimeStr[12:14])
    us = int(usStr)

    dt = datetime(year, month, day, hour, minute, second, us)
    dt = get_timezone().localize(dt)
    return camIdx, dt.astimezone(pytz.utc)


def parse_image_fname_iso(fname):
    basename = os.path.basename(fname)
    name_splitted = basename.split('.')

    def parse_name(name):
        _, camIdxStr, iso_str = name.split('_')
        dt = iso8601.parse_date(iso_str)
        return int(camIdxStr), dt

    try:
        name = '.'.join(name_splitted[:2])
        camIdx, dt = parse_name(name)
    except iso8601.ParseError:
        name = name_splitted[0]
        camIdx, dt = parse_name(name)

    return camIdx, dt


def parse_image_fname(fname, format='auto'):
    if format == 'beesbook':
        return parse_image_fname_beesbook(fname)
    elif format == 'iso':
        return parse_image_fname_iso(fname)
    elif format == 'auto':
        basename = os.path.basename(fname)
        if basename.count('_') >= 3:
            return parse_image_fname_beesbook(fname)
        else:
            return parse_image_fname_iso(fname)
    else:
        raise Exception("Unknown format {}.".format(format))


def parse_video_fname(fname, format='auto'):
    def beesbook_parse():
        begin_name, end_name = basename.split('_TO_')
        (camIdx, begin) = parse_image_fname(begin_name, 'beesbook')
        (_, end) = parse_image_fname(end_name, 'beesbook')
        return camIdx, begin, end

    def iso_parse():
        name, _ = os.path.splitext(basename)
        _, camIdx, isotimespan = name.split('_')
        start, end = isotimespan.split('--')
        end = end.rstrip(".bbb")
        return int(camIdx), iso8601.parse_date(start), iso8601.parse_date(end)

    basename = os.path.basename(fname)
    if format == 'beesbook':
        return beesbook_parse()
    elif format == 'iso':
        return iso_parse()
    elif format == 'auto':
        try:
            return beesbook_parse()
        except ValueError:
            return iso_parse()
    else:
        raise ValueError("Unknown format {}.".format(format))


def to_datetime(t):
    if type(t) in (int, float):
        dt = datetime.fromtimestamp(t, tz=pytz.utc)
        return dt
    elif type(t) == datetime:
        return t
    else:
        raise TypeError("Cannot convert {} to datetime".format(t))


def to_timestamp(dt):
    try:
        return dt.timestamp()
    except AttributeError:  # python 2
        utc_naive = dt.replace(tzinfo=None) - dt.utcoffset()
        timestamp = (utc_naive - datetime(1970, 1, 1)).total_seconds()
        return timestamp