tennis-experiment.py

import random
import pandas as pd

# Build data for experiment

def make_experiment(batch_size, group_size):
  imgs = list(range(454, 920))
  groups = []
  for i in range(group_size, len(imgs)):
    group = imgs[i - group_size:i]
    random.shuffle(group)
    groups.append(group)
  variables_df = pd.DataFrame(random.sample(groups, batch_size), columns=['image{}'.format(i + 1) for i in range(group_size)])
  variables_df.to_csv('variables-{}x{}.csv'.format(batch_size, group_size), index=False)
  return variables_df

for gs in range(2, 5):
  make_experiment(100, gs)
  
batch_df = pd.read_csv('batch-100x4.csv')
answers = batch_df['Answer.ordering']

num_fully_in_order = 0
num_pairs_in_order = 0
num_pairs = 0


# Assessing results

def evaluate(batch_file):
  batch_df = pd.read_csv(batch_file)
  answers = batch_df['Answer.ordering']

  num_samples_in_order = 0
  num_samples = 0
  num_pairs_in_order = 0
  num_pairs = 0

  for answer in answers:
    ordering = [int(id_) for id_ in answer.split('tennis')[1:]]

    if ordering == sorted(ordering):
      num_samples_in_order += 1
    num_samples += 1

    num_pairs_in_order += sum(
        1 for id1, id2 in zip(ordering, ordering[1:]) if id1 < id2)
    num_pairs += len(ordering[1:])
    
  return {
      'num_samples_correct': num_samples_in_order,
      'num_samples': num_samples,
      'num_pairs_correct': num_pairs_in_order,
      'num_pairs': num_pairs,
  }

evaluate('batch-100x4.csv')