def test_factor(collection):
# Full-rank dummy-coding, only one unique value in variable
trial_type = collection.variables['trial_type'].clone()
coll = collection.clone()
transform.Factor(coll, 'trial_type', sep='@')
assert 'trial_type@parametric gain' in coll.variables.keys()
pg = coll.variables['trial_type@parametric gain']
assert pg.values.unique() == [1]
assert pg.values.shape == trial_type.values.shape
# Reduced-rank dummy-coding, only one unique value in variable
coll = collection.clone()
transform.Factor(coll, 'trial_type', constraint='mean_zero')
assert 'trial_type.parametric gain' in coll.variables.keys()
pg = coll.variables['trial_type.parametric gain']
assert pg.values.unique() == [1]
assert pg.values.shape == trial_type.values.shape
# full-rank dummy-coding, multiple values
coll = collection.clone()
transform.Factor(coll, 'respnum')
targets = set(['respnum.%d' % d for d in range(0, 5)])
assert not targets - set(coll.variables.keys())
assert all([
set(coll.variables[t].values.unique()) == {0.0, 1.0} for t in targets
])
data = pd.concat([coll.variables[t].values for t in targets],
axis=1,
sort=True)
assert (data.sum(1) == 1).all()
# reduced-rank dummy-coding, multiple values
coll = collection.clone()
transform.Factor(coll, 'respnum', constraint='drop_one')
targets = set(['respnum.%d' % d for d in range(1, 5)])
assert not targets - set(coll.variables.keys())
assert 'respnum.0' not in coll.variables.keys()
assert all([
set(coll.variables[t].values.unique()) == {0.0, 1.0} for t in targets
])
data = pd.concat([coll.variables[t].values for t in targets],
axis=1,
sort=True)
assert set(np.unique(data.sum(1).values.ravel())) == {0., 1.}
# Effect coding, multiple values
coll = collection.clone()
transform.Factor(coll, 'respnum', constraint='mean_zero')
targets = set(['respnum.%d' % d for d in range(1, 5)])
assert not targets - set(coll.variables.keys())
assert 'respnum.0' not in coll.variables.keys()
assert all([
set(coll.variables[t].values.unique()) == {-0.25, 0.0, 1.0}
for t in targets
])
data = pd.concat([coll.variables[t].values for t in targets],
axis=1,
sort=True)
assert set(np.unique(data.sum(1).values.ravel())) == {-1., 1.}
def test_orthogonalize_dense(collection):
transform.Factor(collection, 'trial_type', sep=sep)
sampling_rate = collection.sampling_rate
# Store pre-orth variables needed for tests
pg_pre = collection['trial_type/parametric gain'].to_dense(sampling_rate)
rt = collection['RT'].to_dense(sampling_rate)
# Orthogonalize and store result
transform.Orthogonalize(collection, variables='trial_type/parametric gain',
other='RT', dense=True, groupby=['run', 'subject'])
pg_post = collection['trial_type/parametric gain']
# Verify that the to_dense() calls result in identical indexing
ent_cols = ['subject', 'run']
assert pg_pre.to_df()[ent_cols].equals(rt.to_df()[ent_cols])
assert pg_post.to_df()[ent_cols].equals(rt.to_df()[ent_cols])
vals = np.c_[rt.values, pg_pre.values, pg_post.values]
df = pd.DataFrame(vals, columns=['rt', 'pre', 'post'])
groupby = rt.get_grouper(['run', 'subject'])
pre_r = df.groupby(groupby).apply(lambda x: x.corr().iloc[0, 1])
post_r = df.groupby(groupby).apply(lambda x: x.corr().iloc[0, 2])
assert (pre_r > 0.2).any()
assert (post_r < 0.0001).all()
def test_or(collection):
coll = collection.clone()
transform.Factor(coll, 'respnum')
names = ['respnum.%d' % d for d in range(0, 5)]
transform.Or(coll, names, output='disjunction')
assert (coll.variables['disjunction'].values == 1).all()
coll['copy'] = coll.variables['respnum.0'].clone()
transform.Or(coll, ['respnum.0', 'copy'], output='or')
assert coll.variables['or'].values.astype(float).equals(
coll.variables['respnum.0'].values)
def test_and(collection):
coll = collection.clone()
transform.Factor(coll, 'respnum')
names = ['respnum.%d' % d for d in range(0, 5)]
coll.variables['respnum.0'].onset += 1
# Should fail because I misaligned variable
with pytest.raises(ValueError):
transform.And(coll, names, output='misaligned')
# Should pass because dense is set to True and will align
transform.And(coll, names, output='misaligned', dense=True)
def test_resample_dense(collection):
new_sampling_rate = 50
old_sampling_rate = collection.sampling_rate
upsampling = float(new_sampling_rate) / old_sampling_rate
collection['RT'] = collection['RT'].to_dense(old_sampling_rate)
old_rt = collection['RT'].clone()
collection.resample(new_sampling_rate, in_place=True)
assert math.floor(len(old_rt.values) * upsampling) == len(collection['RT'].values)
# Should work after explicitly converting categoricals
transform.Factor(collection, 'trial_type')
new_sampling_rate2 = 5
upsampling2 = float(new_sampling_rate2) / old_sampling_rate
collection.resample(new_sampling_rate2, force_dense=True, in_place=True)
assert len(old_rt.values) == math.ceil(float(len(collection['parametric gain'].values) / upsampling2))