def was_there_source_two_or(dataset, narrative, _):
questions = [
'Did you [hear,listen to] [a,an] <S1> or [a,an] <S2>?',
'Have you [heard,listened to] [a,an] <S1> or [a,an] <S2>?'
'Did you [hear,listen to] any <S1> or any <S2>?',
'Have you [heard,listened to] any <S1> or any <S2>?',
'Was there a sound [produced,made] by [a,an] <S1> or a sound [produced,made] by [a,an] <S2>?', # noqa: E501
'Were there any sounds [produced,made] by [a,an] <S1> or any sounds [produced,made] by [a,an] <S2>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
event_1 = str(np.random.choice(dataset['events'])) # sample event
source_1 = str(np.random.choice(
dataset['sources'][event_1])) # sample source
lst_events = [e for e in dataset['events'] if e != event_1]
event_2 = str(np.random.choice(lst_events)) # sample event
source_2 = str(np.random.choice(
dataset['sources'][event_2])) # sample source
question = question.replace('<S1>', source_1) # insert source
question = question.replace('<S2>', source_2) # insert source
question = sanitize_question(question) # correct grammar
lst_sources = get_lst_all_sources(dataset, narrative)
answer = 'yes' if (source_1 in lst_sources
or source_2 in lst_sources) else 'no'
return question, answer
def more_than(dataset, narrative, _):
questions = [
'Were there more <S1>s <A1> than <S2>s <A2>?',
'Was the number of [times,instances,occurrences] [a,an] <S1> <A1> more than the number of [times,instances,occurrences] [a,an] <S2> <A2>?', # noqa: E501
'[Comparing,Listening to,Hearing] the sounds of [a,an] <S1> <A1> and [a,an] <S2> <A2>, were there more [times,instances,occurrences] of the former?', # noqa: E501
'[Comparing,Listening to,Hearing] the sounds of [a,an] <S2> <A2> and [a,an] <S1> <A1>, were there more [times,instances,occurrences] of the latter?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
event_1 = str(np.random.choice(lst_events)) # sample event
source_1 = str(np.random.choice(dataset['sources'][event_1]))
action_1 = str(np.random.choice(dataset['actions'][event_1]))
x_lst_events = [e for e in lst_events if e != event_1]
assert len(x_lst_events) > 0, 'Question (more_than) illposed.'
event_2 = str(np.random.choice(x_lst_events)) # sample event
source_2 = str(np.random.choice(dataset['sources'][event_2]))
action_2 = str(np.random.choice(dataset['actions'][event_2]))
assert event_1 != event_2, 'Question (more_than) illposed.'
question = question.replace('<S1>', source_1) # insert source
question = question.replace('<A1>', action_1) # insert action
question = question.replace('<S2>', source_2) # insert source
question = question.replace('<A2>', action_2) # insert action
question = sanitize_question(question)
answer = 'yes' \
if lst_events.count(event_1) > lst_events.count(event_2) \
else 'no'
return question, answer
def equal_to(dataset, narrative, _):
questions = [
'Was the number of times [a,an] <S1> <A1> equal to the number of times [a,an] <S2> <A2>?', # noqa: E501
'Was the number of times [a,an] <S1> <A1> the same as the number of times [a,an] <S2> <A2>?', # noqa: E501
'Was there an equal number of times [a,an] <S1> <A1> and [a,an] <S2> <A2>?', # noqa: E501
'Was there the same number of <S1> <A1> and <S2> <A2>?',
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
event_1 = str(np.random.choice(lst_events)) # sample event
source_1 = str(np.random.choice(dataset['sources'][event_1]))
action_1 = str(np.random.choice(dataset['actions'][event_1]))
x_lst_events = [e for e in lst_events if e != event_1]
assert len(x_lst_events) > 0, 'Question (equal_to) illposed.'
event_2 = str(np.random.choice(x_lst_events)) # sample event
source_2 = str(np.random.choice(dataset['sources'][event_2]))
action_2 = str(np.random.choice(dataset['actions'][event_2]))
assert event_1 != event_2, 'Question (equal_to) illposed.'
question = question.replace('<S1>', source_1) # insert source
question = question.replace('<A1>', action_1) # insert action
question = question.replace('<S2>', source_2) # insert source
question = question.replace('<A2>', action_2) # insert action
question = sanitize_question(question)
answer = 'yes' \
if lst_events.count(event_1) == lst_events.count(event_2) \
else 'no'
return question, answer
def compare_same_duration_ordinal(dataset, narrative, rel_diff=0.1):
questions = [
'Was the <O1> [sound event,sound] [roughly,approximately] as <D> as the <O2> [sound event,sound]?', # noqa: E501
'Was the <O1> and <O2> [sound events,sounds] [roughly,approximately] as <D>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> [sound event,sound] and the <O2> [sound event,sound], were they [roughly,approximately] as <D>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> [sound event,sound] and the <O2> [sound event,sound], did they [roughly,approximately] have the same duration?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> and <O2> [sound events,sounds], were they [roughly,approximately] as <D>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> and <O2> [sound events,sounds], did they [roughly,approximately] have the same duration?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number_1, ordinal_1 = sample_number(len(lst_events))
duration = sample_duration()
number_2, ordinal_2 = sample_second_number(len(lst_events), number_1)
assert number_1 != number_2, 'Question (compare_same_duration_ordinal) illposed.'
question = question.replace('<O1>', ordinal_1) # insert ordinal
question = question.replace('<D>', duration) # insert duration
question = question.replace('<O2>', ordinal_2) # insert ordinal
question = sanitize_question(question) # correct grammar
lst_duration = get_lst_durations(narrative)
e_1_duration = lst_duration[number_1 - 1]
e_2_duration = lst_duration[number_2 - 1]
rel_duration_diff = compute_rel_diff(np.array(e_1_duration),
np.array(e_2_duration))
# Assert a good margin in relative duration
assert np.sum(np.logical_and(rel_duration_diff > rel_diff,
rel_duration_diff < (2 * rel_diff))) <= 0, \
'Question (compare_same_duration_ordinal) illposed.'
answer = 'yes' if rel_duration_diff <= rel_diff else 'no'
return question, answer
def what_was_loudness(dataset, narrative, rel_diff=0.1):
questions = [
'What was the <AL> sound?',
'What was the <AL> sound you [heard,listened to]?',
'What was the <AL> sound that you [heard,listened to]?',
'What was the <AL> sound that was heard?',
]
question = str(np.random.choice(questions)) # sample question
loudness = sample_absolute_loudness()
question = question.replace('<AL>', loudness) # insert loudness
question = sanitize_question(question) # correct grammar
lst_events = get_lst_events(narrative)
lst_loudness = get_lst_loudness(narrative)
if 'loud' in question:
est = np.argmax(lst_loudness)
elif 'quiet' in question:
est = np.argmin(lst_loudness)
else:
assert False, \
'Loudness illdefined in Question (what_was_loudness).'
# Assert a good margin in relative loudness
evt_loudness = lst_loudness[est]
x_loudness = [j for i, j in enumerate(lst_loudness) if i != est]
rel_loudness_diff = compute_rel_diff(np.array(x_loudness),
np.array(evt_loudness))
assert np.sum(rel_loudness_diff < rel_diff) <= 0, \
'Question (what_was_loudness) illposed.'
e = lst_events[est]
answer = (str(np.random.choice(dataset['sources'][e])) + ' ' +
str(np.random.choice(dataset['actions'][e])))
return question, answer
def how_many_event_two(dataset, narrative, _):
questions = ['How many times was [a,an] <S1> <A1> [or,and] [a,an] <S2> <A2>?',
'How many times did you [hear,listen to] [a,an] <S1> <A1> [or,and] [a,an] <S2> <A2>?', # noqa: E501
'How many times have you [heard,listened to] [a,an] <S1> <A1> [or,and] [a,an] <S2> <A2>?', # noqa: E501
'What is the number of times [a,an] <S1> <A1> [or,and] [a,an] <S2> <A2>?', # noqa: E501
'What is the number of times did you [hear,listen to] [a,an] <S1> <A1> [or,and] [a,an] <S2> <A2>?', # noqa: E501
'What is the number of times you [heard,listened to] [a,an] <S1> <A1> [or,and] [a,an] <S2> <A2>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
event_1 = str(np.random.choice(dataset['events'])) # sample event
source_1 = str(np.random.choice(dataset['sources'][event_1]))
action_1 = str(np.random.choice(dataset['actions'][event_1]))
x_lst_events = [e for e in dataset['events'] if e != event_1]
event_2 = str(np.random.choice(x_lst_events)) # sample event
source_2 = str(np.random.choice(dataset['sources'][event_2]))
action_2 = str(np.random.choice(dataset['actions'][event_2]))
assert event_1 != event_2, 'Question (how_many_event_two) illposed.'
question = question.replace('<S1>', source_1) # insert source
question = question.replace('<A1>', action_1) # insert action
question = question.replace('<S2>', source_2) # insert source
question = question.replace('<A2>', action_2) # insert action
question = sanitize_question(question) # correct grammar
lst_events = get_lst_events(narrative)
answer = numbers_to_words(lst_events.count(event_1)
+ lst_events.count(event_2))
return question, answer
def what_was_duration_relative(dataset, narrative, rel_diff=0.1):
questions = [
'What was the <AD> sound <RO> the <S> <A>?',
'What was the <AD> sound <RO> [hearing,listening to] the <S> <A>?',
'What was the <AD> sound <RO> the <S> <A> was heard?',
]
question = str(np.random.choice(questions)) # sample question
duration = sample_absolute_duration()
preposition = sample_preposition()
lst_events = get_lst_events(narrative)
unique_lst_events = [e for e in lst_events if lst_events.count(e) == 1]
assert len(unique_lst_events) > 0, \
'Question (what_was_duration_relative) illposed.'
event = str(np.random.choice(unique_lst_events))
source = str(np.random.choice(dataset['sources'][event])) # sample source
action = str(np.random.choice(dataset['actions'][event])) # sample action
question = question.replace('<AD>', duration) # insert duration
question = question.replace('<RO>', preposition) # insert preposition
question = question.replace('<S>', source) # insert source
question = question.replace('<A>', action) # insert action
question = sanitize_question(question) # correct grammar
assert lst_events.count(event) == 1, \
'Question (what_was_duration_relative) illposed.'
lst_durations = get_lst_durations(narrative)
event_idx = lst_events.index(event)
if 'before' in question:
lst_events_e = lst_events[:event_idx]
lst_events_d = lst_durations[:event_idx]
elif 'after' in question:
lst_events_e = lst_events[(event_idx + 1):]
lst_events_d = lst_durations[(event_idx + 1):]
else:
assert False, \
'Preposition illdefined in Question (what_was_duration_relative).'
assert len(lst_events_e) > 0, \
'Question (what_was_duration_relative) illposed.'
if 'long' in question:
est = np.argmax(lst_events_d)
elif 'short' in question:
est = np.argmin(lst_events_d)
else:
assert False, \
'Duration illdefined in Question (what_was_duration_relative).'
# Assert a good margin in relative duration
evt_duration = lst_events_d[est]
x_durations = [j for i, j in enumerate(lst_events_d) if i != est]
rel_duration_diff = compute_rel_diff(np.array(x_durations),
np.array(evt_duration))
assert np.sum(rel_duration_diff < rel_diff) <= 0, \
'Question (what_was_duration_relative) illposed.'
e = lst_events_e[est]
answer = (str(np.random.choice(dataset['sources'][e])) + ' ' +
str(np.random.choice(dataset['actions'][e])))
return question, answer
def what_was_duration_relative_ordinal(dataset, narrative, rel_diff=0.1):
questions = [
'What was the <AD> sound <RO> the <O> sound?',
'What was the <AD> sound <RO> [hearing,listening to] the <O> sound?',
'What was the <AD> sound <RO> the <O> sound was heard?',
]
question = str(np.random.choice(questions)) # sample question
duration = sample_absolute_duration()
preposition = sample_preposition()
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<AD>', duration) # insert duration
question = question.replace('<RO>', preposition) # insert preposition
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
lst_durations = get_lst_durations(narrative)
event_idx = (number - 1)
answer = None
if 'before' in question:
if (event_idx - 1) < 0:
answer = 'nothing'
else:
lst_events_e = lst_events[:event_idx]
lst_events_d = lst_durations[:event_idx]
elif 'after' in question:
if (event_idx + 1) >= len(lst_events):
answer = 'nothing'
else:
lst_events_e = lst_events[(event_idx + 1):]
lst_events_d = lst_durations[(event_idx + 1):]
else:
assert False, \
'Preposition illdefined in Question (what_was_duration_relative_ordinal).'
if answer is None:
assert len(lst_events_e) > 0, \
'Question (what_was_duration_relative_ordinal) illposed.'
if 'long' in question:
est = np.argmax(lst_events_d)
elif 'short' in question:
est = np.argmin(lst_events_d)
else:
assert False, \
'Duration illdefined in Question (what_was_duration_relative_ordinal).'
# Assert a good margin in relative duration
evt_duration = lst_events_d[est]
x_durations = [j for i, j in enumerate(lst_events_d) if i != est]
rel_duration_diff = compute_rel_diff(np.array(x_durations),
np.array(evt_duration))
assert np.sum(rel_duration_diff < rel_diff) <= 0, \
'Question (what_was_duration_relative_ordinal) illposed.'
e = lst_events_e[est]
answer = (str(np.random.choice(dataset['sources'][e])) + ' ' +
str(np.random.choice(dataset['actions'][e])))
return question, answer
def was_there_immediate_relative(dataset, narrative, _):
questions = [
'Did you [hear,listen to] [a,an] <S1> <A1> <IO> the <S2> <A2>?', # noqa: E501
'Have you [heard,listened to] [a,an] <S1> <A1> <IO> the <S2> <A2>?', # noqa: E501
'Did you [hear,listen to] any <S1> <A1> <IO> the <S2> <A2>?',
'Have you [heard,listened to] any <S1> <A1> <IO> the <S2> <A2>?',
'Was there [a,an] <S1> <A1> <IO> the <S2> <A2>?',
'Were there any <S1>s <A1> <IO> the <S2> <A2>?',
'Did you [hear,listen to] a sound that [sounds like,sounded like,is,was] [a,an] <S1> <A1> <IO> the <S2> <A2>?', # noqa: E501
'<IO> the <S2> <A2>, did you [hear,listen to] [a,an] <S1> <A1> ?', # noqa: E501
'<IO> the <S2> <A2>, did you [hear,listen to] any <S1> <A1>?',
'<IO> the <S2> <A2>, was there [a,an] <S1> <A1>?',
'<IO> the <S2> <A2>, were there any <S1>s <A1>?',
'<IO> the <S2> <A2>, did you [hear,listen to] a sound that [sounds like,sounded like,is,was] [a,an] <S1> <A1>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
event_1 = str(np.random.choice(dataset['events'])) # sample event
source_1 = str(np.random.choice(dataset['sources'][event_1]))
action_1 = str(np.random.choice(dataset['actions'][event_1]))
preposition = sample_immediate_preposition()
lst_events = get_lst_events(narrative)
unique_lst_events = [e for e in lst_events if lst_events.count(e) == 1]
unique_lst_events = [e for e in unique_lst_events if e != event_1]
assert len(unique_lst_events) > 0, \
'Question (was_there_immediate_relative) illposed.'
event_2 = str(np.random.choice(unique_lst_events))
source_2 = str(np.random.choice(dataset['sources'][event_2]))
action_2 = str(np.random.choice(dataset['actions'][event_2]))
question = question.replace('<S1>', source_1) # insert source
question = question.replace('<A1>', action_1) # insert action
question = question.replace('<IO>', preposition) # insert preposition
question = question.replace('<S2>', source_2) # insert source
question = question.replace('<A2>', action_2) # insert action
question = sanitize_question(question) # correct grammar
assert lst_events.count(event_2) == 1, \
'Question (was_there_immediate_relative) illposed.'
event_2_idx = lst_events.index(event_2)
if 'before' in preposition:
if (event_2_idx - 1) < 0:
target_event = []
else:
target_event = lst_events[event_2_idx - 1]
elif 'after' in preposition:
if (event_2_idx + 1) >= len(lst_events):
target_event = []
else:
target_event = lst_events[event_2_idx + 1]
else:
assert False, \
'Preposition illdefined in Question (was_there_immediate_relative).'
answer = 'yes' if event_1 == target_event else 'no'
return question, answer
def what_was_relative(dataset, narrative, _):
questions = [
'What was the sound <RO> the <S> <A>?',
'What was the sound <RO> [hearing,listening to] the <S> <A>?',
'What was the sound <RO> the <S> <A> was heard?',
'What did you [hear,listen to] <RO> the <S> <A>?',
'What did you [hear,listen to] <RO> [hearing,listening to] the <S> <A>?', # noqa: E501
'What did you [hear,listen to] <RO> the <S> <A> was heard?',
'What was the sound <IO> the <S> <A>?',
'What was the sound <IO> [hearing,listening to] the <S> <A>?',
'What was the sound <IO> the <S> <A> was heard?',
'What did you [hear,listen to] <IO> the <S> <A>?',
'What did you [hear,listen to] <IO> [hearing,listening to] the <S> <A>?', # noqa: E501
'What did you [hear,listen to] <IO> the <S> <A> was heard?',
]
question = str(np.random.choice(questions)) # sample question
preposition = sample_preposition()
immediate_preposition = sample_immediate_preposition()
lst_events = get_lst_events(narrative)
unique_lst_events = [e for e in lst_events if lst_events.count(e) == 1]
assert len(unique_lst_events) > 0, \
'Question (what_was_relative) illposed.'
event = str(np.random.choice(unique_lst_events))
source = str(np.random.choice(dataset['sources'][event])) # sample source
action = str(np.random.choice(dataset['actions'][event])) # sample action
# Only one of the following two lines will have an effect
question = question.replace('<RO>', preposition) # insert preposition
question = question.replace('<IO>', immediate_preposition)
question = question.replace('<S>', source) # insert source
question = question.replace('<A>', action) # insert action
question = sanitize_question(question) # correct grammar
assert lst_events.count(event) == 1, \
'Question (what_was_relative) illposed.'
event_idx = lst_events.index(event)
if 'before' in question:
if (event_idx - 1) < 0:
answer = 'nothing'
else:
e = lst_events[event_idx - 1]
answer = (str(np.random.choice(dataset['sources'][e])) + ' ' +
str(np.random.choice(dataset['actions'][e])))
elif 'after' in question:
if (event_idx + 1) >= len(lst_events):
answer = 'nothing'
else:
e = lst_events[event_idx + 1]
answer = (str(np.random.choice(dataset['sources'][e])) + ' ' +
str(np.random.choice(dataset['actions'][e])))
else:
assert False, 'Preposition illdefined in Question (what_was_relative).'
return question, answer
def compare_duration(dataset, narrative, rel_diff=0.1):
questions = [
'Was the <S1> <A1> <RD> than the <S2> <A2>?',
'Was the sound of the <S1> <A1> <RD> than the sound of the <S2> <A2>?', # noqa: E501
'[Comparing,Listening to,Hearing] the sound of the <S1> <A1> and the sound of the <S2> <A2>, was the former <RD>?', # noqa: E501
'[Comparing,Listening to,Hearing] the sounds of the <S1> <A1> and the <S2> <A2>, was the former <RD>?', # noqa: E501
'[Comparing,Listening to,Hearing] the sound of the <S2> <A2> and the sound of the <S1> <A1>, was the latter <RD>?', # noqa: E501
'[Comparing,Listening to,Hearing] the sounds of the <S2> <A2> and the <S1> <A1>, was the latter <RD>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
unique_lst_events = [e for e in lst_events if lst_events.count(e) == 1]
assert len(unique_lst_events) > 0, \
'Question (compare_duration) illposed.'
event_1 = str(np.random.choice(unique_lst_events)) # sample event
source_1 = str(np.random.choice(dataset['sources'][event_1]))
action_1 = str(np.random.choice(dataset['actions'][event_1]))
rel_duration = sample_rel_duration()
x_unique_lst_events = [e for e in unique_lst_events if e != event_1]
assert len(x_unique_lst_events) > 0, \
'Question (compare_duration) illposed.'
event_2 = str(np.random.choice(x_unique_lst_events)) # sample event
source_2 = str(np.random.choice(dataset['sources'][event_2]))
action_2 = str(np.random.choice(dataset['actions'][event_2]))
assert lst_events.count(event_1) == 1, \
'Question (compare_duration) illposed.'
assert lst_events.count(event_2) == 1, \
'Question (compare_duration) illposed.'
assert event_1 != event_2, 'Question (compare_duration) illposed.'
question = question.replace('<S1>', source_1) # insert source
question = question.replace('<A1>', action_1) # insert action
question = question.replace('<RD>', rel_duration) # insert duration
question = question.replace('<S2>', source_2) # insert source
question = question.replace('<A2>', action_2) # insert action
question = sanitize_question(question)
lst_duration = get_lst_durations(narrative)
e_1_duration = lst_duration[lst_events.index(event_1)]
e_2_duration = lst_duration[lst_events.index(event_2)]
# Assert a good margin in relative duration
rel_duration_diff = compute_rel_diff(np.array(e_1_duration),
np.array(e_2_duration))
assert np.sum(rel_duration_diff < rel_diff) <= 0, \
'Question (compare_duration) illposed.'
if 'short' in question:
answer = 'yes' if e_1_duration < e_2_duration else 'no'
elif 'long' in question:
answer = 'yes' if e_1_duration > e_2_duration else 'no'
else:
assert False, 'Duration illdefined in Question (compare_duration).'
return question, answer
def compare_same_duration(dataset, narrative, rel_diff=0.1):
questions = [
'Was the <S1> <A1> [roughly,approximately] as <D> as the <S2> <A2>?', # noqa: E501
'Was the sound of the <S1> <A1> [roughly,approximately] as <D> as the sound of the <S2> <A2>?', # noqa: E501
'Was the sound of the <S1> <A1> [roughly,approximately] the same duration as the sound of the <S2> <A2>?', # noqa: E501
'[Comparing,Listening to,Hearing] the sound of the <S1> <A1> and the sound of the <S2> <A2>, did they [roughly,approximately] have the same duration?', # noqa: E501
'[Comparing,Listening to,Hearing] the sounds of the <S1> <A1> and the <S2> <A2>, did they [roughly,approximately] have the same duration?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
unique_lst_events = [e for e in lst_events if lst_events.count(e) == 1]
assert len(unique_lst_events) > 0, \
'Question (compare_same_duration) illposed.'
event_1 = str(np.random.choice(unique_lst_events)) # sample event
source_1 = str(np.random.choice(dataset['sources'][event_1]))
action_1 = str(np.random.choice(dataset['actions'][event_1]))
duration = sample_duration()
x_unique_lst_events = [e for e in unique_lst_events if e != event_1]
assert len(x_unique_lst_events) > 0, \
'Question (compare_same_duration) illposed.'
event_2 = str(np.random.choice(x_unique_lst_events)) # sample event
source_2 = str(np.random.choice(dataset['sources'][event_2]))
action_2 = str(np.random.choice(dataset['actions'][event_2]))
assert lst_events.count(event_1) == 1, \
'Question (compare_same_duration) illposed.'
assert lst_events.count(event_2) == 1, \
'Question (compare_same_duration) illposed.'
assert event_1 != event_2, 'Question (compare_same_duration) illposed.'
question = question.replace('<S1>', source_1) # insert source
question = question.replace('<A1>', action_1) # insert action
question = question.replace('<D>', duration) # insert duration
question = question.replace('<S2>', source_2) # insert source
question = question.replace('<A2>', action_2) # insert action
question = sanitize_question(question)
lst_duration = get_lst_durations(narrative)
e_1_duration = lst_duration[lst_events.index(event_1)]
e_2_duration = lst_duration[lst_events.index(event_2)]
rel_duration_diff = compute_rel_diff(np.array(e_1_duration),
np.array(e_2_duration))
# Assert a good margin in relative duration
assert np.sum(np.logical_and(rel_duration_diff > rel_diff,
rel_duration_diff < (2 * rel_diff))) <= 0, \
'Question (compare_same_duration) illposed.'
answer = 'yes' if rel_duration_diff <= rel_diff else 'no'
return question, answer
def compare_same_duration_event_ordinal(dataset, narrative, rel_diff=0.1):
questions = [
'Was the <S> <A> [roughly,approximately] as <D> as the <O> [sound event,sound]?', # noqa: E501
'[Comparing,Listening to,Hearing] the <S> <A> and the <O> [sound event,sound], were they [roughly,approximately] as <D>?', # noqa: E501
'[Comparing,Listening to,Hearing] the sound of the <S> <A> and the <O> [sound event,sound], were they [roughly,approximately] as <D>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <S> <A> and the <O> [sound event,sound], did they [roughly,approximately] have the same duration?', # noqa: E501
'[Comparing,Listening to,Hearing] the sound of the <S> <A> and the <O> [sound event,sound], did they [roughly,approximately] have the same duration?', # noqa: E501
'Was the <O> [sound event,sound] [roughly,approximately] as <D> as the <S> <A>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O> [sound event,sound] and the <S> <A>, were they [roughly,approximately] as <D>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O> [sound event,sound] and the sound of the <S> <A>, were they [roughly,approximately] as <D>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O> [sound event,sound] and the <S> <A>, did they [roughly,approximately] have the same duration?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O> [sound event,sound] and the sound of the <S> <A>, did they [roughly,approximately] have the same duration?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
unique_lst_events = [e for e in lst_events if lst_events.count(e) == 1]
assert len(unique_lst_events) > 0, \
'Question (compare_same_duration_event_ordinal) illposed.'
event = str(np.random.choice(unique_lst_events)) # sample event
source = str(np.random.choice(dataset['sources'][event]))
action = str(np.random.choice(dataset['actions'][event]))
duration = sample_duration()
number, ordinal = sample_second_number(len(lst_events),
lst_events.index(event) + 1)
assert lst_events.count(event) == 1, \
'Question (compare_same_duration_event_ordinal) illposed.'
assert lst_events.index(event) != (number - 1), \
'Question (compare_same_duration_event_ordinal) illposed.'
question = question.replace('<S>', source) # insert source
question = question.replace('<A>', action) # insert action
question = question.replace('<D>', duration) # insert duration
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
lst_duration = get_lst_durations(narrative)
e_1_duration = lst_duration[lst_events.index(event)]
e_2_duration = lst_duration[number - 1]
rel_duration_diff = compute_rel_diff(np.array(e_1_duration),
np.array(e_2_duration))
# Assert a good margin in relative duration
assert np.sum(np.logical_and(rel_duration_diff > rel_diff,
rel_duration_diff < (2 * rel_diff))) <= 0, \
'Question (compare_same_duration_event_ordinal) illposed.'
answer = 'yes' if rel_duration_diff <= rel_diff else 'no'
return question, answer
def how_many_sounds_duration_event(dataset, narrative, rel_diff=0.1):
questions = ['How many [sound events,sounds] [roughly,approximately] as <D> as the <S> <A>?', # noqa: E501
'How many [sound events,sounds] that are [roughly,approximately] as <D> as the <S> <A> [did,could] you [hear,listen to]?', # noqa: E501
'How many [sound events,sounds] that are [roughly,approximately] as <D> as the <S> <A> have you heard?', # noqa: E501
'How many [sound events,sounds] that have [roughly,approximately] the same duration as the <S> <A>?', # noqa: E501
'How many [sound events,sounds] that have [roughly,approximately] the same duration as the <S> <A> [did,could] you [hear,listen to]?', # noqa: E501
'How many [sound events,sounds] that have [roughly,approximately] the same duration as the <S> <A> have you heard?', # noqa: E501
'What is the number of [sound events,sounds] [roughly,approximately] as <D> as the <S> <A>?', # noqa: E501
'What is the number of [sound events,sounds] that are [roughly,approximately] as <D> as the <S> <A> [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of [sound events,sounds] that are [roughly,approximately] as <D> as the <S> <A> have you heard?', # noqa: E501
'What is the number of [sound events,sounds] that have [roughly,approximately] the same duration as the <S> <A>?', # noqa: E501
'What is the number of [sound events,sounds] that have [roughly,approximately] the same duration as the <S> <A> [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of [sound events,sounds] that have [roughly,approximately] the same duration as the <S> <A> have you heard?', # noqa: E501
'There is [a,an] <S> <A>; how many [sound events,sounds] that are [roughly,approximately] as <D>?', # noqa: E501
'There is [a,an] <S> <A>; what is the number of [sound events,sounds] that are [roughly,approximately] as <D>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
duration = sample_duration() # sample duration
lst_events = get_lst_events(narrative)
unique_lst_events = [e for e in lst_events if lst_events.count(e) == 1]
assert len(unique_lst_events) > 0, \
'Question (how_many_sounds_duration_event) illposed.'
event = str(np.random.choice(unique_lst_events))
source = str(np.random.choice(dataset['sources'][event])) # sample source
action = str(np.random.choice(dataset['actions'][event])) # sample action
question = question.replace('<D>', duration) # insert duration
question = question.replace('<S>', source) # insert source
question = question.replace('<A>', action) # insert action
question = sanitize_question(question) # correct grammar
assert lst_events.count(event) == 1, \
'Question (how_many_sounds_duration_event) illposed.'
lst_durations = get_lst_durations(narrative)
event_idx = lst_events.index(event)
evt_duration = lst_durations[event_idx]
x_durations = [j for i, j in enumerate(lst_durations) if i != event_idx]
rel_durations_diff = compute_rel_diff(np.array(x_durations),
np.array(evt_duration))
# Assert a good margin in relative loudness
assert np.sum(np.logical_and(rel_durations_diff > rel_diff,
rel_durations_diff < (2 * rel_diff))) <= 0, \
'Question (how_many_sounds_duration_event) illposed.'
answer = numbers_to_words(np.sum(rel_durations_diff <= rel_diff))
return question, answer
def compare_duration_ordinal_event(dataset, narrative, rel_diff=0.1):
questions = [
'Was the <O> [sound event,sound] <RD> than the <S> <A>?',
'Was the <O> [sound event,sound] <RD> than the sound of the <S> <A>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O> [sound event,sound] and the <S> <A>, was the former <RD>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <S> <A> and the <O> [sound event,sound], was the latter <RD>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
unique_lst_events = [e for e in lst_events if lst_events.count(e) == 1]
assert len(unique_lst_events) > 0, \
'Question (compare_duration_ordinal_event) illposed.'
event = str(np.random.choice(unique_lst_events)) # sample event
source = str(np.random.choice(dataset['sources'][event]))
action = str(np.random.choice(dataset['actions'][event]))
rel_duration = sample_rel_duration()
number, ordinal = sample_second_number(len(lst_events),
lst_events.index(event) + 1)
assert lst_events.count(event) == 1, \
'Question (compare_duration_ordinal_event) illposed.'
assert lst_events.index(event) != (number - 1), \
'Question (compare_duration_ordinal_event) illposed.'
question = question.replace('<S>', source) # insert source
question = question.replace('<A>', action) # insert action
question = question.replace('<RD>', rel_duration) # insert duration
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
lst_duration = get_lst_durations(narrative)
e_1_duration = lst_duration[number - 1]
e_2_duration = lst_duration[lst_events.index(event)]
# Assert a good margin in relative duration
rel_duration_diff = compute_rel_diff(np.array(e_1_duration),
np.array(e_2_duration))
assert np.sum(rel_duration_diff < rel_diff) <= 0, \
'Question (compare_duration_ordinal_event) illposed.'
if 'short' in question:
answer = 'yes' if e_1_duration < e_2_duration else 'no'
elif 'long' in question:
answer = 'yes' if e_1_duration > e_2_duration else 'no'
else:
assert False, \
'Duration illdefined in Question (compare_duration_ordinal_event).'
return question, answer
def was_there_similar_duration(dataset, narrative, rel_diff=0.1):
questions = [
'Were there any sounds [roughly,approximately] as <D> as the <S> <A>?', # noqa: E501
'Were there any sounds that were [roughly,approximately] as <D> as the <S> <A>?', # noqa: E501
'Were there any sounds that were [roughly,approximately] the same duration as the <S> <A>?', # noqa: E501
'Was there any sound [roughly,approximately] as <D> as the <S> <A>?', # noqa: E501
'Was there any sound that was [roughly,approximately] as <D> as the <S> <A>?', # noqa: E501
'Was there any sound that was [roughly,approximately] the same duration as the <S> <A>?', # noqa: E501
'Was there at least a sound [roughly,approximately] as <D> as the <S> <A>?', # noqa: E501
'Was there at least a sound that was [roughly,approximately] as <D> as the <S> <A>?', # noqa: E501
'Was there at least a sound that was [roughly,approximately] the same duration as <S> <A>?', # noqa: E501
'Was there at least [one,a single] sound [roughly,approximately] as <D> as the <S> <A>?', # noqa: E501
'Was there at least [one,a single] sound that was [roughly,approximately] as <D> as the <S> <A>?', # noqa: E501
'Was there at least [one,a single] sound that was [roughly,approximately] the same duration as <S> <A>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
duration = sample_duration() # sample duration
lst_events = get_lst_events(narrative)
unique_lst_events = [e for e in lst_events if lst_events.count(e) == 1]
assert len(unique_lst_events) > 0, \
'Question (was_there_similar_duration) illposed.'
event = str(np.random.choice(unique_lst_events))
source = str(np.random.choice(dataset['sources'][event])) # sample source
action = str(np.random.choice(dataset['actions'][event])) # sample action
question = question.replace('<D>', duration) # insert duration
question = question.replace('<S>', source) # insert source
question = question.replace('<A>', action) # insert action
question = sanitize_question(question) # correct grammar
assert lst_events.count(event) == 1, \
'Question (was_there_similar_duration) illposed.'
lst_durations = get_lst_durations(narrative)
event_idx = lst_events.index(event)
evt_duration = lst_durations[event_idx]
x_durations = [j for i, j in enumerate(lst_durations) if i != event_idx]
rel_durations_diff = compute_rel_diff(np.array(x_durations),
np.array(evt_duration))
# Assert a good margin in relative duration
assert np.sum(np.logical_and(rel_durations_diff > rel_diff,
rel_durations_diff < (2 * rel_diff))) <= 0, \
'Question (was_there_similar_duration) illposed.'
answer = 'yes' if np.sum(rel_durations_diff <= rel_diff) >= 1 else 'no'
return question, answer
def how_many_event_relative(dataset, narrative, _):
questions = ['How many <S1>s <A1> <RO> the <S2> <A2> were there?',
'How many <S1>s <A1> <RO> the <S2> <A2> [did,could] you [hear,listen to]?', # noqa: E501
'How many <S1>s <A1> <RO> the <S2> <A2> have you [heard,listened to]?', # noqa: E501
'What is the number of <S1>s <A1> <RO> the <S2> <A2>?',
'What is the number of <S1>s <A1> <RO> the <S2> <A2> [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of <S1>s <A1> <RO> the <S2> <A2> have you [heard,listened to]?', # noqa: E501
'There is [a,an] <S2> <A2>; how many <S1>s <A1> [did,could] you hear <RO>?', # noqa: E501
'There is [a,an] <S2> <A2>; how many <S1>s <A1> have you heard <RO>?', # noqa: E501
'There is [a,an] <S2> <A2>; what is the number of <S1>s <A1> [did,could] you hear <RO>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
event_1 = str(np.random.choice(dataset['events'])) # sample event
source_1 = str(np.random.choice(dataset['sources'][event_1]))
action_1 = str(np.random.choice(dataset['actions'][event_1]))
preposition = sample_preposition()
lst_events = get_lst_events(narrative)
unique_lst_events = [e for e in lst_events if lst_events.count(e) == 1]
unique_lst_events = [e for e in unique_lst_events if e != event_1]
assert len(unique_lst_events) > 0, \
'Question (how_many_event_relative) illposed.'
event_2 = str(np.random.choice(unique_lst_events))
source_2 = str(np.random.choice(dataset['sources'][event_2]))
action_2 = str(np.random.choice(dataset['actions'][event_2]))
question = question.replace('<S1>', source_1) # insert source
question = question.replace('<A1>', action_1) # insert action
question = question.replace('<RO>', preposition) # insert preposition
question = question.replace('<S2>', source_2) # insert source
question = question.replace('<A2>', action_2) # insert action
question = sanitize_question(question) # correct grammar
assert lst_events.count(event_2) == 1, \
'Question (how_many_event_relative) illposed.'
event_2_idx = lst_events.index(event_2)
if 'before' in question:
lst_events_e = lst_events[:event_2_idx]
elif 'after' in question:
lst_events_e = lst_events[(event_2_idx + 1):]
else:
assert False, \
'Relative preposition illdefined in Question (how_many_event_relative).'
answer = numbers_to_words(lst_events_e.count(event_1))
return question, answer
def how_many(dataset, narrative, _):
questions = ['How many [sound events,sounds] were there?',
'How many [sound events,sounds] [did,could] you [hear,listen to]?',
'How many [sound events,sounds] have you [heard,listened to]?',
'What is the number of [sound events,sounds]?',
'What is the number of [sound events,sounds] [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of [sound events,sounds] have you [heard,listened to]?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
question = sanitize_question(question) # correct grammar
lst_events = get_lst_events(narrative)
answer = numbers_to_words(len(lst_events))
return question, answer
def how_many_ordinal(dataset, narrative, _):
questions = ['How many times did you [hear,listen to] a sound that [sounded,seemed] like the <O> [sound event,sound]?', # noqa: E501
'What is the number of times did you [hear,listen to] a sound that [sounded,seemed] like the <O> [sound event,sound]?', # noqa: E501
'[Hearing,Listening to] the <O> [sound event,sound], how many sounds were [the same, similar]?', # noqa: E501
'[Hearing,Listening to] the <O> [sound event,sound], what is the number of sounds that were [the same, similar]?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
event = lst_events[number - 1]
answer = numbers_to_words(lst_events.count(event) - 1) # -1 for base event
return question, answer
def what_was(dataset, narrative, _):
questions = [
'What was the <O> sound you [heard,listened to]?',
'What was the <O> sound?',
'What did the <O> sound [sound,seem] like?',
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
event = lst_events[number - 1]
answer = (str(np.random.choice(dataset['sources'][event])) + ' ' +
str(np.random.choice(dataset['actions'][event])))
return question, answer
def how_many_sounds_relative(dataset, narrative, _):
questions = ['How many [sound events,sounds] <RO> the <S> <A> were there?',
'How many [sound events,sounds] <RO> the <S> <A> [did,could] you [hear,listen to]?', # noqa: E501
'How many [sound events,sounds] <RO> the <S> <A> have you [heard,listened to]?', # noqa: E501
'What is the number of [sound events,sounds] <RO> the <S> <A>?',
'What is the number of [sound events,sounds] <RO> the <S> <A> [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of [sound events,sounds] <RO> the <S> <A> have you [heard,listened to]?', # noqa: E501
'There is [a,an] <S> <A>; how many [sound events,sounds] [did,could] you hear <RO>?', # noqa: E501
'There is [a,an] <S> <A>; how many [sound events,sounds] have you heard <RO>?', # noqa: E501
'There is [a,an] <S> <A>; what is the number of [sound events,sounds] [did,could] you hear <RO>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
preposition = sample_preposition()
lst_events = get_lst_events(narrative)
unique_lst_events = [e for e in lst_events if lst_events.count(e) == 1]
assert len(unique_lst_events) > 0, \
'Question (how_many_sounds_relative) illposed.'
event = str(np.random.choice(unique_lst_events))
source = str(np.random.choice(dataset['sources'][event])) # sample source
action = str(np.random.choice(dataset['actions'][event])) # sample action
question = question.replace('<RO>', preposition) # insert preposition
question = question.replace('<S>', source) # insert source
question = question.replace('<A>', action) # insert action
question = sanitize_question(question) # correct grammar
assert lst_events.count(event) == 1, \
'Question (how_many_sounds_relative) illposed.'
event_idx = lst_events.index(event)
if 'before' in question:
lst_events_e = lst_events[:event_idx]
elif 'after' in question:
lst_events_e = lst_events[(event_idx + 1):]
else:
assert False, \
'Preposition illdefined in Question (how_many_sounds_relative).'
answer = numbers_to_words(len(lst_events_e))
return question, answer
def was_there_similar_ordinal(dataset, narrative, _):
questions = [
'Were there any similar sounds to the <O> sound?',
'Were there any sounds that were similar to the <O> sound?',
'Was there at least a sound similar to the <O> sound?',
'Was there at least a sound that was similar to the <O> sound?', # noqa: E501
'Was there at least [one,a single] sound similar to the <O> sound?',
'Was there at least [one,a single] sound that was similar to the <O> sound?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
event = lst_events[number - 1]
answer = 'yes' if lst_events.count(event) > 1 else 'no' # 1 for reference
return question, answer
def was_there_source(dataset, narrative, _):
questions = [
'Did you [hear,listen to] [a,an] <S>?',
'Have you [heard,listened to] [a,an] <S>?'
'Did you [hear,listen to] any <S>?',
'Have you [heard,listened to] any <S>?',
'Was there a sound [produced,made] by [a,an] <S>?',
'Were there any sounds [produced,made] by [a,an] <S>?',
]
question = str(np.random.choice(questions)) # sample question
event = str(np.random.choice(dataset['events'])) # sample event
source = str(np.random.choice(dataset['sources'][event])) # sample source
question = question.replace('<S>', source) # insert source
question = sanitize_question(question) # correct grammar
answer = 'yes' if source in get_lst_all_sources(dataset,
narrative) else 'no'
return question, answer
def how_many_sounds_relative_ordinal(dataset, narrative, _):
questions = ['How many [sound events,sounds] after the <O> [sound event,sound] were there?', # noqa: E501
'How many [sound events,sounds] after the <O> [sound event,sound] [did,could] you [hear,listen to]?', # noqa: E501
'How many [sound events,sounds] after the <O> [sound event,sound] have you [heard,listened to]?', # noqa: E501
'What is the number of [sound events,sounds] after the <O> [sound event,sound]?', # noqa: E501
'What is the number of [sound events,sounds] after the <O> [sound event,sound] [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of [sound events,sounds] after the <O> [sound event,sound] have you [heard,listened to]?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
assert number < (len(lst_events) - 1), \
'Question (how_many_sounds_relative_ordinal) illposed.'
lst_events_e = lst_events[number:]
answer = numbers_to_words(len(lst_events_e))
return question, answer
def how_many_event(dataset, narrative, _):
questions = ['How many times was [a,an] <S> <A>?',
'How many times did you [hear,listen to] [a,an] <S> <A>?',
'How many times have you [heard,listened to] [a,an] <S> <A>?',
'What is the number of times [a,an] <S> <A>?',
'What is the number of times did you [hear,listen to] [a,an] <S> <A>?', # noqa: E501
'What is the number of times you [heard,listened to] [a,an] <S> <A>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
event = str(np.random.choice(lst_events))
source = str(np.random.choice(dataset['sources'][event])) # sample source
action = str(np.random.choice(dataset['actions'][event])) # sample action
question = question.replace('<S>', source) # insert source
question = question.replace('<A>', action) # insert action
question = sanitize_question(question) # correct grammar
answer = numbers_to_words(lst_events.count(event))
return question, answer
def compare_ordinal(dataset, narrative, _):
questions = [
'Was the <O1> [sound event,sound] [the same as,similar to] the <O2> [sound event,sound]?', # noqa: E501
'Was the <O1> [sound event,sound] and <O2> [sound event,sound] [the same,similar]?', # noqa: E501
'Were the <O1> and <O2> [sound events,sounds] [the same,similar]?',
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number_1, ordinal_1 = sample_number(len(lst_events))
number_2, ordinal_2 = sample_second_number(len(lst_events), number_1)
assert number_1 != number_2, 'Question (compare_ordinal) illposed.'
question = question.replace('<O1>', ordinal_1) # insert ordinal
question = question.replace('<O2>', ordinal_2) # insert ordinal
question = sanitize_question(question) # correct grammar
answer = 'yes' if lst_events[number_1 - 1] == lst_events[number_2 - 1] \
else 'no'
return question, answer
def was_there_two_or(dataset, narrative, _):
questions = [
'Did you [hear,listen to] [a,an] <S1> <A1> or [a,an] <S2> <A2>?',
'Have you [heard,listened to] [a,an] <S1> <A1> or [a,an] <S2> <A2>?',
'Did you [hear,listen to] any <S1> <A1> or any <S2> <A2>?',
'Have you [heard,listened to] any <S1> <A1> or any <S2> <A2>?',
'Did you [hear,listen to] a sound that [sounds like,is] [a,an] <S1> <A1> or a sound [sounds like,is] [a,an] <S2> <A2>?', # noqa: E501
'Did you [hear,listen to] a sound that [sounded like,was] [a,an] <S1> <A1> or a sound [sounded like,was] [a,an] <S2> <A2>?', # noqa: E501
'Have you [heard,listened to] a sound that [sounds like,is] [a,an] <S1> <A1> or a sound [sounds like,is] [a,an] <S2> <A2>?', # noqa: E501
'Have you [heard,listened to] a sound that [sounded like,was] [a,an] <S1> <A1> or a sound [sounded like,was] [a,an] <S2> <A2>?', # noqa: E501
'Was there [a,an] <S1> <A1> or [a,an] <S2> <A2>?',
'Were there any <S1>s <A1> or any <S2>s <A2>?',
]
question = str(np.random.choice(questions)) # sample question
event_1 = str(np.random.choice(dataset['events'])) # sample event
source_1 = str(np.random.choice(
dataset['sources'][event_1])) # sample source
action_1 = str(np.random.choice(
dataset['actions'][event_1])) # sample action
lst_events = [e for e in dataset['events'] if e != event_1]
event_2 = str(np.random.choice(lst_events)) # sample event
source_2 = str(np.random.choice(
dataset['sources'][event_2])) # sample source
action_2 = str(np.random.choice(
dataset['actions'][event_2])) # sample action
question = question.replace('<S1>', source_1) # insert source
question = question.replace('<A1>', action_1) # insert action
question = question.replace('<S2>', source_2) # insert source
question = question.replace('<A2>', action_2) # insert action
question = sanitize_question(question) # correct grammar
lst_events = get_lst_events(narrative)
answer = 'yes' if (event_1 in lst_events
or event_2 in lst_events) else 'no'
return question, answer
def how_many_sounds_duration_ordinal(dataset, narrative, rel_diff=0.1):
questions = ['How many [sound events,sounds] [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'How many [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'How many [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound [did,could] you [hear,listen to]?', # noqa: E501
'How many [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound have you heard?', # noqa: E501
'How many [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound?', # noqa: E501
'How many [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound [did,could] you [hear,listen to]?', # noqa: E501
'How many [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound have you heard?', # noqa: E501
'What is the number of [sound events,sounds] [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'What is the number of [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'What is the number of [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound have you heard?', # noqa: E501
'What is the number of [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound?', # noqa: E501
'What is the number of [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound have you heard?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
duration = sample_duration() # sample duration
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<D>', duration) # insert duration
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
lst_durations = get_lst_durations(narrative)
evt_duration = lst_durations[number - 1]
x_durations = [j for i, j in enumerate(lst_durations) if i != (number - 1)]
rel_durations_diff = compute_rel_diff(np.array(x_durations),
np.array(evt_duration))
# Assert a good margin in relative loudness
assert np.sum(np.logical_and(rel_durations_diff > rel_diff,
rel_durations_diff < (2 * rel_diff))) <= 0, \
'Question (how_many_sounds_duration_ordinal) illposed.'
answer = numbers_to_words(np.sum(rel_durations_diff <= rel_diff))
return question, answer
def compare_duration_ordinal(dataset, narrative, rel_diff=0.1):
questions = [
'Was the <O1> [sound event,sound] <RD> than the <O2> [sound event,sound]?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> [sound event,sound] and the <O2> [sound event,sound], was the former <RD>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> and <O2> [sound events,sounds], was the former <RD>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O2> [sound event,sound] and the <O1> [sound event,sound], was the latter <RD>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O2> and <O1> [sound events,sounds], was the latter <RD>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number_1, ordinal_1 = sample_number(len(lst_events))
rel_duration = sample_rel_duration()
number_2, ordinal_2 = sample_second_number(len(lst_events), number_1)
assert number_1 != number_2, 'Question (compare_duration_ordinal) illposed.'
question = question.replace('<O1>', ordinal_1) # insert ordinal
question = question.replace('<RD>', rel_duration) # insert duration
question = question.replace('<O2>', ordinal_2) # insert ordinal
question = sanitize_question(question) # correct grammar
lst_duration = get_lst_durations(narrative)
e_1_duration = lst_duration[number_1 - 1]
e_2_duration = lst_duration[number_2 - 1]
# Assert a good margin in relative duration
rel_duration_diff = compute_rel_diff(np.array(e_1_duration),
np.array(e_2_duration))
assert np.sum(rel_duration_diff < rel_diff) <= 0, \
'Question (compare_duration_ordinal) illposed.'
if 'short' in question:
answer = 'yes' if e_1_duration < e_2_duration else 'no'
elif 'long' in question:
answer = 'yes' if e_1_duration > e_2_duration else 'no'
else:
assert False, 'Duration illdefined in Question (compare_duration_ordinal).'
return question, answer
def compare_ordinal_event(dataset, narrative, _):
questions = [
'Was the <O> [sound event,sound] [a,an] <S> <A>?', # noqa: E501
'Did the <O> [sound event,sound] [sound,seem] like [a,an] <S> <A>?', # noqa: E501
'[Listening to,Hearing] the <O> [sound event,sound], was it [a,an] <S> <A>?', # noqa: E501
'[Listening to,Hearing] the <O> [sound event,sound], did it [sound,seem] like [a,an] <S> <A>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
event = str(np.random.choice(dataset['events'])) # sample event
source = str(np.random.choice(dataset['sources'][event])) # sample source
action = str(np.random.choice(dataset['actions'][event])) # sample action
question = question.replace('<O>', ordinal) # insert ordinal
question = question.replace('<S>', source) # insert source
question = question.replace('<A>', action) # insert action
question = sanitize_question(question) # correct grammar
answer = 'yes' if lst_events[number - 1] == event else 'no'
return question, answer
