def __init__(self, identifier):
regression = linear_regression(xarray_kwargs=dict(stimulus_coord='stimulus_id')) # word
correlation = pearsonr_correlation(xarray_kwargs=dict(correlation_coord='stimulus_id'))
metric = CrossRegressedCorrelation(regression=regression, correlation=correlation,
crossvalidation_kwargs=dict(splits=5, kfold=True, split_coord='stimulus_id',
stratification_coord='sentence_id'))
super(Fedorenko2016Encoding, self).__init__(identifier=identifier, metric=metric)
def fROI_correlation():
assembly = load_voxels()
stories = list(sorted(set(assembly['story'].values)))
subjects = list(sorted(set(assembly['subject_UID'].values)))
split_scores = []
correlate = pearsonr_correlation(xarray_kwargs=dict(correlation_coord='stimulus_id', neuroid_coord='fROI_area'))
cross_stories_subjects = list(itertools.product(stories, subjects))
for story, heldout_subject in tqdm(cross_stories_subjects, desc='cross-{story,subject}'):
story_assembly = assembly[{'presentation': [coord_story == story for coord_story in assembly['story'].values]}]
subject_pool = story_assembly[{'neuroid': [subject != heldout_subject
for subject in story_assembly['subject_UID'].values]}]
subject_pool = average_subregions(subject_pool)
heldout = story_assembly[{'neuroid': [subject == heldout_subject
for subject in story_assembly['subject_UID'].values]}]
heldout = average_subregions(heldout)
split_score = correlate(subject_pool, heldout)
split_score = type(split_score)(split_score.values, coords={
coord: (dims, values) for coord, dims, values in walk_coords(split_score)
if not coord.startswith('subject_') and coord != 'neuroid_id'}, dims=split_score.dims)
split_score = split_score.expand_dims('heldout_subject').expand_dims('story')
split_score['heldout_subject'], split_score['story'] = [heldout_subject], [story]
split_scores.append(split_score)
correlation = Score.merge(*split_scores)
correlation = apply_aggregate(lambda scores: scores.mean('neuroid').mean('story'), correlation)
center = correlation.mean('heldout_subject')
error = correlation.std('heldout_subject')
score = Score([center, error], coords={**{'aggregation': ['center', 'error']},
**{coord: (dims, values) for coord, dims, values in walk_coords(center)}},
dims=('aggregation',) + center.dims)
score.attrs[Score.RAW_VALUES_KEY] = correlation.attrs[Score.RAW_VALUES_KEY]
return score
def ToliasCadena2017PLS():
loader = AssemblyLoader()
assembly_repetition = loader(average_repetition=False)
assembly = loader(average_repetition=True)
assembly.stimulus_set.name = assembly.stimulus_set_name
similarity_metric = CrossRegressedCorrelation(
regression=pls_regression(),
correlation=pearsonr_correlation(),
crossvalidation_kwargs={'stratification_coord': None})
identifier = f'tolias.Cadena2017-pls'
ceiler = InternalConsistency(split_coord='repetition_id')
def ceiling():
# This assembly has many stimuli that are only shown to a subset of the neurons.
# When the loader runs with `average_repetition=True`, it automatically drops nan cells,
# but for the `assembly_repetition`, it keeps all the rows.
# If we now use the usual ceiling approach, the two halves will end up with NaN values
# due to stimuli not being shown to neurons, which doesn't let us correlate.
# Instead, we here drop all NaN cells and their corresponding stimuli,
# which keeps only 43% of the original presentation rows, but lets us correlate again.
assembly_nonan, stimuli = loader.dropna(assembly_repetition, assembly_repetition.attrs['stimulus_set'])
return ceiler(assembly_nonan)
return NeuralBenchmark(identifier=identifier, version=1,
assembly=assembly, similarity_metric=similarity_metric, visual_degrees=VISUAL_DEGREES,
ceiling_func=ceiling)
def __init__(self,
identifier,
split_coord='word',
unique_split_values=False):
self._logger = logging.getLogger(fullname(self))
self._identifier = identifier
assembly = LazyLoad(self._load_assembly)
self._target_assembly = assembly
regression = linear_regression(xarray_kwargs=dict(
stimulus_coord='word_id',
neuroid_coord='subject_id')) # used for sorting -- keep at word_id
correlation = pearsonr_correlation(xarray_kwargs=dict(
correlation_coord='word_id',
neuroid_coord='subject_id')) # used for sorting -- keep at word_id
self._metric = CrossRegressedCorrelation(
regression=regression,
correlation=correlation,
crossvalidation_kwargs=dict(
splits=5,
kfold=True,
split_coord=split_coord,
stratification_coord='sentence_id',
unique_split_values=unique_split_values))
self._cross_subject = CartesianProduct(dividers=['subject_id'])
self._ceiler = self.ManySubjectExtrapolationCeiling(
subject_column='subject_id')
def __init__(self):
ceiling = Score([1, np.nan],
coords={'aggregation': ['center', 'error']},
dims=['aggregation'])
assembly_repetition = get_assembly()
assert len(np.unique(assembly_repetition['region'])) == 1
assert hasattr(assembly_repetition, 'repetition')
self.region = 'IT'
self.assembly = average_repetition(assembly_repetition)
self._assembly = self.assembly
self.timebins = timebins_from_assembly(self.assembly)
self._similarity_metric = CrossRegressedCorrelation(
regression=pls_regression(),
correlation=pearsonr_correlation(),
crossvalidation_kwargs=dict(
stratification_coord=Split.Defaults.stratification_coord
if hasattr(self.assembly, Split.Defaults.stratification_coord
) else None))
identifier = f'{assembly_repetition.name}-layer_selection'
ceiler = InternalConsistency()
super(_MockBenchmark,
self).__init__(identifier=identifier,
ceiling_func=lambda: ceiler(assembly_repetition),
version='1.0')
def __init__(self, **kwargs):
metric = CrossRegressedCorrelation(
regression=linear_regression(xarray_kwargs=dict(stimulus_coord='stimulus_id')),
correlation=pearsonr_correlation(xarray_kwargs=dict(correlation_coord='stimulus_id')),
crossvalidation_kwargs=dict(split_coord='stimulus_id', stratification_coord=None))
metric = Invert(metric)
super(PereiraDecoding, self).__init__(metric=metric, **kwargs)
def DicarloMajaj2015V4MaskParams(*regression_args, **regression_kwargs):
return _DicarloMajaj2015Region('V4', identifier_metric_suffix='param_mask',
similarity_metric=ScaledCrossRegressedCorrelation(
regression=mask_regression_with_params(
*regression_args, **regression_kwargs),
correlation=pearsonr_correlation(),
crossvalidation_kwargs=dict(splits=4, stratification_coord='object_name')),
ceiler=InternalConsistency())
def DicarloSanghavi2020ITPLS():
return _DicarloSanghavi2020Region(
'IT',
identifier_metric_suffix='pls',
similarity_metric=CrossRegressedCorrelation(
regression=pls_regression(),
correlation=pearsonr_correlation(),
crossvalidation_kwargs=dict(stratification_coord='object_name')),
ceiler=InternalConsistency())
def MovshonFreemanZiemba2013V2PLS():
return _MovshonFreemanZiemba2013Region(
'V2',
identifier_metric_suffix='pls',
similarity_metric=CrossRegressedCorrelation(
regression=pls_regression(),
correlation=pearsonr_correlation(),
crossvalidation_kwargs=dict(stratification_coord='texture_type')),
ceiler=InternalConsistency())
def __init__(self, bold_shift=None):
assembly = load_Fedorenko2016()
self._target_assembly = assembly
self._regression = pls_regression(xarray_kwargs=dict(stimulus_coord='stimulus_id')) # word
self._correlation = pearsonr_correlation(xarray_kwargs=dict(correlation_coord='stimulus_id'))
self._metric = CrossRegressedCorrelation(
regression=self._regression, correlation=self._correlation,
# crossvalidation_kwargs=dict(split_coord='sentence_id', stratification_coord=None))
crossvalidation_kwargs=dict(split_coord='stimulus_id', stratification_coord='sentence_id', train_size=.8))
def DicarloMajaj2015ITMask():
return _DicarloMajaj2015Region(
'IT',
identifier_metric_suffix='mask',
similarity_metric=ScaledCrossRegressedCorrelation(
regression=mask_regression(),
correlation=pearsonr_correlation(),
crossvalidation_kwargs=dict(splits=2,
stratification_coord='object_name')),
ceiler=InternalConsistency())
def __init__(self, *args, **kwargs):
super(Blank2014fROIEncoding, self).__init__(*args, **kwargs)
regression = linear_regression(xarray_kwargs=dict(
stimulus_coord='stimulus_id', neuroid_coord='fROI_area'))
correlation = pearsonr_correlation(xarray_kwargs=dict(
correlation_coord='stimulus_id', neuroid_coord='fROI_area'))
self._metric = CrossRegressedCorrelation(
regression=regression, correlation=correlation,
crossvalidation_kwargs=dict(splits=5, kfold=True, split_coord='stimulus_id', stratification_coord='story'))
def test_small(self):
assembly = NeuroidAssembly((np.arange(30 * 25) + np.random.standard_normal(30 * 25)).reshape((30, 25)),
coords={'image_id': ('presentation', np.arange(30)),
'object_name': ('presentation', ['a', 'b', 'c'] * 10),
'neuroid_id': ('neuroid', np.arange(25)),
'region': ('neuroid', ['some_region'] * 25)},
dims=['presentation', 'neuroid'])
metric = CrossRegressedCorrelation(regression=pls_regression(), correlation=pearsonr_correlation())
score = metric(source=assembly, target=assembly)
assert score.sel(aggregation='center') == approx(1, abs=.00001)
def DicarloMajaj2015TransRegHighVar(*regression_args, **regression_kwargs):
return _DicarloMajaj2015RegHighVar(
identifier_metric_suffix='var6_trans',
coord_list=['ty', 'tz'],
similarity_metric=ScaledCrossRegressedCorrelation(
regression=XarrayRegression(
MultiOutputRegressor(LinearSVR(
*regression_args, **regression_kwargs))),
correlation=pearsonr_correlation(),
crossvalidation_kwargs=dict(
splits=4, stratification_coord='object_name')))
def DicarloMajaj2015RotSemRegLowMidVar(*regression_args, **regression_kwargs):
return _DicarloMajaj2015RegLowMidVar(
identifier_metric_suffix='var03_rotsem',
coord_list=['rxy_semantic', 'ryz_semantic', 'rxz_semantic'],
similarity_metric=ScaledCrossRegressedCorrelation(
regression=XarrayRegression(
MultiOutputRegressor(LinearSVR(
*regression_args, **regression_kwargs))),
correlation=pearsonr_correlation(),
crossvalidation_kwargs=dict(
splits=4, stratification_coord='object_name')))
def __init__(self, bold_shift=None):
assembly = load_Fedorenko2016()
self._target_assembly = assembly
# avg code
# packaging file, change the benchmark to include averaging.
self._regression = pls_regression(xarray_kwargs=dict(stimulus_coord='stimulus_id')) # word
self._correlation = pearsonr_correlation(xarray_kwargs=dict(correlation_coord='stimulus_id'))
self._metric = CrossRegressedCorrelation(
regression=self._regression, correlation=self._correlation,
crossvalidation_kwargs=dict(split_coord='stimulus_id', stratification_coord='stimulus_id'))
def _standard_benchmark(identifier, load_assembly, visual_degrees, number_of_trials, stratification_coord, bibtex):
assembly_repetition = LazyLoad(lambda: load_assembly(average_repetitions=False))
assembly = LazyLoad(lambda: load_assembly(average_repetitions=True))
similarity_metric = CrossRegressedCorrelation(
regression=pls_regression(), correlation=pearsonr_correlation(),
crossvalidation_kwargs=dict(stratification_coord=stratification_coord))
ceiler = InternalConsistency()
return NeuralBenchmark(identifier=f"{identifier}-pls", version=1,
assembly=assembly, similarity_metric=similarity_metric,
visual_degrees=visual_degrees, number_of_trials=number_of_trials,
ceiling_func=lambda: ceiler(assembly_repetition),
parent=None,
bibtex=bibtex)
def __init__(self, identifier, bold_shift=4):
self._identifier = identifier
assembly = LazyLoad(lambda: self._load_assembly(bold_shift))
self._target_assembly = assembly
regression = linear_regression(xarray_kwargs=dict(
stimulus_coord='stimulus_id', neuroid_coord='neuroid_id'))
correlation = pearsonr_correlation(xarray_kwargs=dict(
correlation_coord='stimulus_id', neuroid_coord='neuroid_id'))
self._metric = CrossRegressedCorrelation(
regression=regression, correlation=correlation,
crossvalidation_kwargs=dict(splits=5, kfold=True, split_coord='stimulus_id', stratification_coord='story'))
self._ceiler = ExtrapolationCeiling(subject_column='subject_UID', post_process=self.post_process_ceilings)
def __init__(self, *args, **kwargs):
super(Futrell2018SentencesEncoding, self).__init__(*args, **kwargs)
regression = linear_regression(xarray_kwargs=dict(
stimulus_coord='word_id', neuroid_coord='subject_id'))
correlation = pearsonr_correlation(xarray_kwargs=dict(
correlation_coord='word_id', neuroid_coord='subject_id'))
self._metric = CrossRegressedCorrelation(
regression=regression,
correlation=correlation,
crossvalidation_kwargs=dict(splits=5,
kfold=True,
unique_split_values=True,
split_coord='sentence_id',
stratification_coord=None))
def ToliasCadena2017Mask():
loader = AssemblyLoader()
assembly_repetition = loader(average_repetition=False)
assembly = loader(average_repetition=True)
assembly.stimulus_set.name = assembly.stimulus_set_name
similarity_metric = CrossRegressedCorrelation(
regression=mask_regression(),
correlation=pearsonr_correlation(),
crossvalidation_kwargs={'splits': 4, 'stratification_coord': None})
identifier = f'tolias.Cadena2017-mask'
ceiler = InternalConsistency(split_coord='repetition_id')
return NeuralBenchmark(identifier=identifier, version=1,
assembly=assembly, similarity_metric=similarity_metric, visual_degrees=VISUAL_DEGREES,
ceiling_func=lambda: ceiler(assembly_repetition))
def __init__(self, identifier):
self._logger = logging.getLogger(fullname(self))
self._identifier = identifier
assembly = LazyLoad(self._load_assembly)
self._target_assembly = assembly
regression = linear_regression(xarray_kwargs=dict(
stimulus_coord='word_id', neuroid_coord='subject_id'))
correlation = pearsonr_correlation(xarray_kwargs=dict(
correlation_coord='word_id', neuroid_coord='subject_id'))
self._metric = CrossRegressedCorrelation(
regression=regression,
correlation=correlation,
crossvalidation_kwargs=dict(splits=5,
kfold=True,
split_coord='word_id',
stratification_coord='sentence_id'))
def _standard_benchmark(identifier, load_assembly, stratification_coord):
assembly_repetition = LazyLoad(
lambda: load_assembly(average_repetitions=False))
assembly = LazyLoad(lambda: load_assembly(average_repetitions=True))
similarity_metric = CrossRegressedCorrelation(
regression=pls_regression(),
correlation=pearsonr_correlation(),
crossvalidation_kwargs=dict(stratification_coord=stratification_coord))
ceiler = InternalConsistency()
return NeuralBenchmark(
identifier=f"{identifier}-pls",
version=1,
assembly=assembly,
similarity_metric=similarity_metric,
ceiling_func=lambda: ceiler(assembly_repetition),
parent=None,
paper_link='http://www.jneurosci.org/content/35/39/13402.short')
def test_across_images(self):
values = (np.arange(30 * 25 * 5) +
np.random.standard_normal(30 * 25 * 5)).reshape((30, 25, 5))
assembly = NeuroidAssembly(
values,
coords={
'image_id': ('presentation', np.arange(30)),
'object_name': ('presentation', ['a', 'b', 'c'] * 10),
'neuroid_id': ('neuroid', np.arange(25)),
'region': ('neuroid', ['some_region'] * 25),
'time_bin_start': ('time_bin', list(range(5))),
'time_bin_end': ('time_bin', list(range(1, 6))),
},
dims=['presentation', 'neuroid', 'time_bin'])
correlation = TemporalCorrelationAcrossImages(pearsonr_correlation())
score = correlation(assembly, assembly)
np.testing.assert_array_equal(score.dims, ['neuroid'])
np.testing.assert_array_equal(score['neuroid_id'].values,
list(range(25)))
np.testing.assert_array_almost_equal(score.values, [1.] * 25)
assert set(score.raw.dims) == {'neuroid', 'time_bin'}
def ToliasCadena2017MaskParams(*regression_args, **regression_kwargs):
loader = AssemblyLoader()
assembly_repetition = loader(average_repetition=False)
assembly = loader(average_repetition=True)
assembly.stimulus_set.name = assembly.stimulus_set_name
similarity_metric = CrossRegressedCorrelation(
regression=mask_regression_with_params(
*regression_args, **regression_kwargs),
correlation=pearsonr_correlation(),
crossvalidation_kwargs={'splits': 4, 'stratification_coord': None})
identifier = f'tolias.Cadena2017-param-mask'
ceiler = InternalConsistency(split_coord='repetition_id')
def ceiling():
assembly_nonan, stimuli = loader.dropna(
assembly_repetition,
assembly_repetition.attrs['stimulus_set'])
return ceiler(assembly_nonan)
return NeuralBenchmark(identifier=identifier, version=1,
assembly=assembly, similarity_metric=similarity_metric,
ceiling_func=ceiling)
