def test_export_model_to_hdf5(file_stimuli, tmpdir):
model = pysaliency.UniformModel()
filename = str(tmpdir.join('model.hdf5'))
export_model_to_hdf5(model, file_stimuli, filename)
model2 = pysaliency.HDF5Model(file_stimuli, filename)
for s in file_stimuli:
np.testing.assert_allclose(model.log_density(s), model2.log_density(s))
def test_model(self):
stimuli = pysaliency.Stimuli([np.random.randn(40, 40, 3),
np.random.randn(40, 50, 3)])
model = pysaliency.UniformModel()
np.testing.assert_allclose(model.log_likelihoods(stimuli, self.f),
[-np.log(40*40),
-np.log(40*40),
-np.log(40*40),
-np.log(40*40),
-np.log(40*40),
-np.log(40*50),
-np.log(40*50),
-np.log(40*50),
])
def test_crossval_splits(many_stimuli, crossval_folds, val_folds, test_folds):
if not test_folds and val_folds != 1:
return # this case is raises an implementation error right now
tmp_model = pysaliency.UniformModel()
fixations = tmp_model.sample(many_stimuli, 100)
train_stimuli = []
train_fixations = []
val_stimuli = []
val_fixations = []
test_stimuli = []
test_fixations = []
for _train_stimuli, _train_fixations, _val_stimuli, _val_fixations, _test_stimuli, _test_fixations in \
filter_datasets.iterate_crossvalidation(many_stimuli, fixations, crossval_folds=crossval_folds, val_folds=val_folds, test_folds=test_folds, random=True):
assert not set(_train_stimuli.stimulus_ids).intersection(
_val_stimuli.stimulus_ids)
assert not set(_train_stimuli.stimulus_ids).intersection(
_test_stimuli.stimulus_ids)
if not test_folds: # otherwise test is validation
assert not set(_val_stimuli.stimulus_ids).intersection(
_test_stimuli.stimulus_ids)
train_stimuli.append(_train_stimuli)
train_fixations.append(_train_fixations)
val_stimuli.append(_val_stimuli)
val_fixations.append(_val_fixations)
test_stimuli.append(_test_stimuli)
test_fixations.append(_test_fixations)
assert sum(len(s) for s in val_stimuli) == val_folds * len(many_stimuli)
assert sum(len(s) for s in test_stimuli) == test_folds * len(many_stimuli)
assert sum(len(s)
for s in train_stimuli) == (crossval_folds - val_folds -
test_folds) * len(many_stimuli)
assert sum(len(f.x) for f in val_fixations) == val_folds * len(fixations.x)
assert sum(len(f.x)
for f in test_fixations) == test_folds * len(fixations.x)
assert sum(len(f.x)
for f in train_fixations) == (crossval_folds - val_folds -
test_folds) * len(fixations.x)
assert len(train_stimuli) == crossval_folds
def test_stratified_crossval_splits_multiple_attributes(
many_stimuli, crossval_folds, val_folds, test_folds):
if not test_folds and val_folds != 1:
return # this case is raises an implementation error right now
tmp_model = pysaliency.UniformModel()
fixations = tmp_model.sample(many_stimuli, 100)
train_stimuli = []
train_fixations = []
val_stimuli = []
val_fixations = []
test_stimuli = []
test_fixations = []
for _train_stimuli, _train_fixations, _val_stimuli, _val_fixations, _test_stimuli, _test_fixations in \
filter_datasets.iterate_crossvalidation(many_stimuli, fixations, crossval_folds=crossval_folds, val_folds=val_folds, test_folds=test_folds, random=True, stratified_attributes=['category', 'category2']):
assert not set(_train_stimuli.stimulus_ids).intersection(
_val_stimuli.stimulus_ids)
assert not set(_train_stimuli.stimulus_ids).intersection(
_test_stimuli.stimulus_ids)
if not test_folds: # otherwise test is validation
assert not set(_val_stimuli.stimulus_ids).intersection(
_test_stimuli.stimulus_ids)
np.testing.assert_allclose(
np.sum(_train_stimuli.attributes['category'] == 0),
len(many_stimuli) / crossval_folds *
(crossval_folds - val_folds - test_folds) * 0.1,
atol=1)
np.testing.assert_allclose(
np.sum(_val_stimuli.attributes['category'] == 0),
len(many_stimuli) / crossval_folds * val_folds * 0.1,
atol=1)
np.testing.assert_allclose(
np.sum(_test_stimuli.attributes['category'] == 0),
len(many_stimuli) / crossval_folds * test_folds * 0.1,
atol=1)
train_stimuli.append(_train_stimuli)
train_fixations.append(_train_fixations)
val_stimuli.append(_val_stimuli)
val_fixations.append(_val_fixations)
test_stimuli.append(_test_stimuli)
test_fixations.append(_test_fixations)
assert sum(len(s) for s in val_stimuli) == val_folds * len(many_stimuli)
assert sum(len(s) for s in test_stimuli) == test_folds * len(many_stimuli)
assert sum(len(s)
for s in train_stimuli) == (crossval_folds - val_folds -
test_folds) * len(many_stimuli)
assert sum(len(f.x) for f in val_fixations) == val_folds * len(fixations.x)
assert sum(len(f.x)
for f in test_fixations) == test_folds * len(fixations.x)
assert sum(len(f.x)
for f in train_fixations) == (crossval_folds - val_folds -
test_folds) * len(fixations.x)
assert len(train_stimuli) == crossval_folds