providence/scripts/import_annotations.py

#!usr/bin/env python
# -*- coding: utf8 -*-

#
# Author: William N. Havard (base on various files by Lucas Gautheron)
#

import os
import argparse
import logging
import pandas as pd
pd.set_option('mode.chained_assignment',None) # Silences pandas' complaints

from ChildProject.projects import ChildProject
from ChildProject.annotations import AnnotationManager

def _raw_filename(path):
    return os.path.splitext(path)[0]

def _filter_missing_annotation_files(input):
    annotation_path = lambda row: os.path.join('annotations', row['set'], 'raw', row['raw_filename'])
    input['exists'] = input.apply(lambda row: os.path.exists(annotation_path(row)), axis=1)
    missing_annotations = input[input['exists'] == False]
    input = input[input['exists'] == True]

    if len(missing_annotations):
        missing_annotations['expected_path'] = missing_annotations.apply(lambda row: annotation_path(row), axis=1)
        missing_annotations_path = sorted(missing_annotations['expected_path'].tolist())
        missing_message = "[WARNING] Some annotations you expected to have are missing.\n"\
                          "Check whether these annotations exist and if so, if their expected path "\
                          "reflect their true path.\n\t - {}".format('\n\t - '.join(missing_annotations_path))
        print(missing_message)
    return input

def _check_importation(am, annotation_set, expected_ann_number):
    annotations = am.annotations[am.annotations['set'] == annotation_set]
    if len(annotations) != expected_ann_number:
        print('[WARNING] Expected to import {} annotations, only found {}!'.
              format(len(annotations), expected_ann_number))

    annotations_segments = am.get_segments(annotations)
    if len(annotations_segments) == 0:
        print('[WARNING] Annotations were imported, but they either contain no segments '
              'or the segments were not imported properly!')

    return len(annotations), len(annotations_segments)

def _get_recordings(project, annotation_set, ann_format):
    input = project.recordings[['recording_filename', 'duration', 'child_id']]
    input.dropna(inplace=True)
    input = input[input['recording_filename'] != 'NA']
    input['set'] = annotation_set
    input['time_seek'] = 0
    input['range_onset'] = 0
    input['range_offset'] = input['duration']
    input['format'] = ann_format
    input.drop(['duration'], axis=1, inplace=True)

    return input

def main(annotation_set):
    annotation_set = annotation_set.lower()
    ann_format = '{}_rttm'.format(annotation_set) if annotation_set in ('vcm', 'vtc') else annotation_set

    # Load project
    project = ChildProject('.')
    am = AnnotationManager(project)

    # Get recordings and set up df
    input = _get_recordings(project, annotation_set, ann_format)

    # RAW FILENAME
    # /!\ input['raw_filename'] refers to the *subpath and name* of the *annotation file* to be imported
    # not the *target* recording file. input['raw_filename'] should hence match the
    # name of the annotation files you want to import! The filenames of the annotation files
    # usually have the same name or contain the name of the recording. Hence, their name depends
    # on the value of input['recording_filename'] (unless in case of bulk importation, see below).
    # (input['raw_filename'] = 'SUB/PATH/annotation_filename.txt')

    # FILTERING
    # /!\ Using input['filter'] to filter the input is not necessary but safer.
    # Filtering the *input annotation file* is necessary in case of bulk importation
    # where the annotation file contains annotations linked to *several recordings*. In such case,
    # it should be specified for each input['recording_filename'] which lines of the annotation file
    # should be associated to it.

    # Set up 'raw_filename' and 'filter' depending on the annotation set to import
    if annotation_set in ('vcm', 'vtc'):
        # My VCM annotations end with '.vcm' and not '.rttm'. Update this if it's not the case for you.
        rttm_ext = 'rttm' if annotation_set == 'vtc' else 'vcm'
        input['raw_filename'] = input['recording_filename'].apply(lambda f: '{}.{}'.format(_raw_filename(f), rttm_ext))
        # We only keep lines for which the 'file' column is equal to filter (unnecessary here, but safer)
        input['filter'] = input['recording_filename'].apply(lambda f: os.path.basename(_raw_filename(f)))
    elif annotation_set == 'alice':
        # My ALICE annotations have the same name as the recording filename. Update if it's not the case for you.
        input['raw_filename'] = input['recording_filename'].apply(lambda f: '{}.{}'.format(_raw_filename(f), 'txt'))
        # We only keep lines for which the 'file' column is equal to filter (unnecessary here, but safer)
        input['filter'] = input['recording_filename'].apply(lambda f: os.path.basename(_raw_filename(f)))
    elif annotation_set == 'cha':
        input['raw_filename'] = input['recording_filename'].apply(lambda f: '{}.{}'.format(_raw_filename(f), 'cha'))
        # CHA files do not need filtering as they only contain annotations for the file they are linked to
    else:
        raise ValueError('Unknown annotation set `{}`!'.format(annotation_set))

    # Filter out rows for which we do not find the matching annotation file
    input = _filter_missing_annotation_files(input)
    expected_ann_number = len(input)

    # Do importation
    if len(input) > 0:
        am.remove_set(annotation_set)
        am.import_annotations(input)
        len_ann, len_seg = _check_importation(am, annotation_set, expected_ann_number)
        print('Imported {} annotations resulting in {} segments!'.format(len_ann, len_seg))
    else:
        print('[WARNING] Nothing to import!')

if __name__ == '__main__':
    accepted_annotations = ['cha', 'vtc', 'vcm', 'alice']

    parser = argparse.ArgumentParser(description='Import {} annotations.'.format('/'.join(accepted_annotations)))
    parser.add_argument('annotation_set',
                        help='Set of annotations to import (amongst {}).'.format('/'.join(accepted_annotations)),
                        choices=accepted_annotations,
                        type= lambda s: s.lower())
    args = parser.parse_args()

    main(args.annotation_set)