Script to read Charades CSV files (Charades_v1_train.csv, Charades_v1_test.csv) into a pandas DataFrame where each row represents a single action

charades_csv_reader.py

import pandas as pd
from csv import DictReader
from collections import defaultdict
import numpy as np
def read_charades_csv(path):
    def parse_action(action_str):
        cls, start, stop = action_str.split(' ')
        return {
            'class': int(cls[1:]),
            'start': float(start),
            'stop': float(stop)
        }

    def parse_actions(actions_str):
        actions = actions_str.split(';')
        return [parse_action(action.strip()) for action in actions]

    def maybe_null(reader_fn):
        def reader(val):
            if isinstance(val, str) and len(val) == 0:
                return np.nan
            return reader_fn(val)
        return reader
                
    readers = {
        'objects': lambda objects_str: objects_str.split(';'),
        'length': maybe_null(float),
        'quality': maybe_null(int),
        'relevance': maybe_null(int),
    }
    with open(path, 'r') as f:
        columns = defaultdict(lambda: [])
        reader = DictReader(f)
        for row in reader:
            cols = list(row.keys())
            if len(row['actions']) > 0:
                actions = parse_actions(row['actions'])
                action_count = len(actions)

                for col in set(cols) - set(['actions']):
                    if col in readers:
                        value = readers[col](row[col])
                    else:
                        value = row[col]
                    columns[col].extend([value] * action_count)
                    
                columns['action'].extend([a['class'] for a in actions])
                columns['start_time'].extend([a['start'] for a in actions])
                columns['stop_time'].extend([a['stop'] for a in actions])
        return pd.DataFrame(dict(columns))