def _pairwise_corr(self, columns=None, covar=None, tail='two-sided',
method='pearson', padjust='none', export_filename=None):
"""Pairwise (partial) correlations."""
stats = pairwise_corr(data=self, columns=columns, covar=covar,
tail=tail, method=method, padjust=padjust,
export_filename=export_filename)
return stats
def test_pairwise_corr(self):
"""Test function pairwise_corr"""
# Load JASP Big 5 DataSets (remove subject column)
data = read_dataset('pairwise_corr').iloc[:, 1:]
stats = pairwise_corr(data=data, method='pearson', tail='two-sided')
jasp_rval = [
-0.350, -0.01, -.134, -.368, .267, .055, .065, .159, -.013, .159
]
assert np.allclose(stats['r'].values, jasp_rval)
assert stats['n'].values[0] == 500
# Correct for multiple comparisons
pairwise_corr(data=data,
method='spearman',
tail='one-sided',
padjust='bonf')
# Export
pairwise_corr(data=data,
method='spearman',
tail='one-sided',
export_filename='test_export.csv')
# Check with a subset of columns
pairwise_corr(data=data, columns=['Neuroticism', 'Extraversion'])
with pytest.raises(ValueError):
pairwise_corr(data=data, tail='wrong')
with pytest.raises(ValueError):
pairwise_corr(data=data, columns='wrong')
# Check with non-numeric columns
data['test'] = 'test'
pairwise_corr(data=data, method='pearson')
# Check different variation of product / combination
n = data.shape[0]
data['Age'] = np.random.randint(18, 65, n)
data['IQ'] = np.random.normal(105, 1, n)
data['One'] = 1
data['Gender'] = np.repeat(['M', 'F'], int(n / 2))
pairwise_corr(data,
columns=['Neuroticism', 'Gender'],
method='shepherd')
pairwise_corr(data, columns=['Neuroticism', 'Extraversion', 'Gender'])
pairwise_corr(data, columns=['Neuroticism'])
pairwise_corr(data, columns='Neuroticism', method='skipped')
pairwise_corr(data, columns=[['Neuroticism']], method='spearman')
pairwise_corr(data, columns=[['Neuroticism'], None], method='percbend')
pairwise_corr(data, columns=[['Neuroticism', 'Gender'], ['Age']])
pairwise_corr(data, columns=[['Neuroticism'], ['Age', 'IQ']])
pairwise_corr(data, columns=[['Age', 'IQ'], []])
pairwise_corr(data, columns=['Age', 'Gender', 'IQ', 'Wrong'])
pairwise_corr(data, columns=['Age', 'Gender', 'Wrong'])
# Test with no good combinations
with pytest.raises(ValueError):
pairwise_corr(data, columns=['Gender', 'Gender'])
# Test when one column has only one unique value
pairwise_corr(data=data, columns=['Age', 'One', 'Gender'])
stats = pairwise_corr(data, columns=['Neuroticism', 'IQ', 'One'])
assert stats.shape[0] == 1
# Test with covariate
pairwise_corr(data, covar='Age')
pairwise_corr(data, covar=['Age', 'Neuroticism'])
with pytest.raises(AssertionError):
pairwise_corr(data, covar=['Age', 'Gender'])
with pytest.raises(ValueError):
pairwise_corr(data, columns=['Neuroticism', 'Age'], covar='Age')
# Partial pairwise with missing values
data.loc[4, 'Age'] = np.nan
data.loc[10, 'Neuroticism'] = np.nan
pairwise_corr(data, covar='Age')
######################################################################
# MultiIndex columns
from numpy.random import random as rdm
# Create MultiIndex dataframe
columns = pd.MultiIndex.from_tuples([('Behavior', 'Rating'),
('Behavior', 'RT'),
('Physio', 'BOLD'),
('Physio', 'HR'),
('Psycho', 'Anxiety')])
data = pd.DataFrame(
dict(Rating=rdm(size=10),
RT=rdm(size=10),
BOLD=rdm(size=10),
HR=rdm(size=10),
Anxiety=rdm(size=10)))
data.columns = columns
pairwise_corr(data, method='spearman')
stats = pairwise_corr(data, columns=[('Behavior', 'Rating')])
assert stats.shape[0] == data.shape[1] - 1
pairwise_corr(data,
columns=[('Behavior', 'Rating'), ('Behavior', 'RT')])
st1 = pairwise_corr(data,
columns=[[('Behavior', 'Rating'),
('Behavior', 'RT')], None])
st2 = pairwise_corr(data,
columns=[[('Behavior', 'Rating'),
('Behavior', 'RT')]])
assert st1['X'].equals(st2['X'])
st3 = pairwise_corr(data,
columns=[[('Behavior', 'Rating')],
[('Behavior', 'RT'), ('Physio', 'BOLD')]])
assert st3.shape[0] == 2
# With covar
pairwise_corr(data, covar=[('Psycho', 'Anxiety')])
pairwise_corr(data,
columns=[('Behavior', 'Rating')],
covar=[('Psycho', 'Anxiety')])
def test_pairwise_corr(self):
"""Test function pairwise_corr"""
# Load JASP Big 5 DataSets (remove subject column)
data = read_dataset('pairwise_corr').iloc[:, 1:]
stats = pairwise_corr(data=data, method='pearson', tail='two-sided')
jasp_rval = [-0.350, -0.01, -.134, -.368, .267, .055, .065, .159,
-.013, .159]
assert np.allclose(stats['r'].values, jasp_rval)
assert stats['n'].values[0] == 500
# Correct for multiple comparisons
pairwise_corr(data=data, method='spearman', tail='one-sided',
padjust='bonf')
# Export
pairwise_corr(data=data, method='spearman', tail='one-sided',
export_filename='test_export.csv')
# Check with a subset of columns
pairwise_corr(data=data, columns=['Neuroticism', 'Extraversion'])
with pytest.raises(ValueError):
pairwise_corr(data=data, tail='wrong')
# Check with non-numeric columns
data['test'] = 'test'
pairwise_corr(data=data, method='pearson')
# Check different variation of product / combination
n = data.shape[0]
data['Age'] = np.random.randint(18, 65, n)
data['IQ'] = np.random.normal(105, 1, n)
data['Gender'] = np.repeat(['M', 'F'], int(n / 2))
pairwise_corr(data, columns=['Neuroticism', 'Gender'])
pairwise_corr(data, columns=['Neuroticism', 'Extraversion', 'Gender'])
pairwise_corr(data, columns=['Neuroticism'])
pairwise_corr(data, columns='Neuroticism')
pairwise_corr(data, columns=[['Neuroticism']])
pairwise_corr(data, columns=[['Neuroticism'], None])
pairwise_corr(data, columns=[['Neuroticism', 'Gender'], ['Age']])
pairwise_corr(data, columns=[['Neuroticism'], ['Age', 'IQ']])
pairwise_corr(data, columns=[['Age', 'IQ'], []])
pairwise_corr(data, columns=['Age', 'Gender', 'IQ', 'Wrong'])
pairwise_corr(data, columns=['Age', 'Gender', 'Wrong'])
# Test with more than 1000 columns (BF10 not computed)
data1500 = pd.concat([data, data, data], ignore_index=True)
pcor1500 = pairwise_corr(data1500, method='pearson')
assert 'BF10' not in pcor1500.keys()
def test_pairwise_corr(self):
"""Test function pairwise_corr"""
# Load JASP Big 5 DataSets (remove subject column)
data = read_dataset('pairwise_corr').iloc[:, 1:]
stats = pairwise_corr(data=data, method='pearson', tail='two-sided')
jasp_rval = [-0.350, -0.01, -.134, -.368, .267, .055, .065, .159,
-.013, .159]
assert np.allclose(stats['r'].round(3).to_numpy(), jasp_rval)
assert stats['n'].to_numpy()[0] == 500
# Correct for multiple comparisons
pairwise_corr(data=data, method='spearman', tail='one-sided',
padjust='bonf')
# Check with a subset of columns
pairwise_corr(data=data, columns=['Neuroticism', 'Extraversion'])
with pytest.raises(ValueError):
pairwise_corr(data=data, columns='wrong')
# Check with non-numeric columns
data['test'] = 'test'
pairwise_corr(data=data, method='pearson')
# Check different variation of product / combination
n = data.shape[0]
data['Age'] = np.random.randint(18, 65, n)
data['IQ'] = np.random.normal(105, 1, n)
data['One'] = 1
data['Gender'] = np.repeat(['M', 'F'], int(n / 2))
pairwise_corr(data, columns=['Neuroticism', 'Gender'],
method='shepherd')
pairwise_corr(data, columns=['Neuroticism', 'Extraversion', 'Gender'])
pairwise_corr(data, columns=['Neuroticism'])
pairwise_corr(data, columns='Neuroticism', method='skipped')
pairwise_corr(data, columns=[['Neuroticism']], method='spearman')
pairwise_corr(data, columns=[['Neuroticism'], None], method='percbend')
pairwise_corr(data, columns=[['Neuroticism', 'Gender'], ['Age']])
pairwise_corr(data, columns=[['Neuroticism'], ['Age', 'IQ']])
pairwise_corr(data, columns=[['Age', 'IQ'], []])
pairwise_corr(data, columns=['Age', 'Gender', 'IQ', 'Wrong'])
pairwise_corr(data, columns=['Age', 'Gender', 'Wrong'])
# Test with no good combinations
with pytest.raises(ValueError):
pairwise_corr(data, columns=['Gender', 'Gender'])
# Test when one column has only one unique value
pairwise_corr(data=data, columns=['Age', 'One', 'Gender'])
stats = pairwise_corr(data, columns=['Neuroticism', 'IQ', 'One'])
assert stats.shape[0] == 1
# Test with covariate
pairwise_corr(data, covar='Age')
pairwise_corr(data, covar=['Age', 'Neuroticism'])
with pytest.raises(AssertionError):
pairwise_corr(data, covar=['Age', 'Gender'])
# Partial pairwise with overlapping covariates
pairwise_corr(
data.drop(columns=['One', 'Gender']),
covar=data.drop(columns=['test', 'One', 'Gender']).columns
)
pairwise_corr(data, columns='Neuroticism', covar='Age')
with pytest.raises(AssertionError):
pairwise_corr(data, columns=['Neuroticism', 'Age'], covar='One')
# Test against pcorr:
_pcorr = data.drop(columns=['One', 'Gender']).pcorr().stack()
_pwcorr = pairwise_corr(
data.drop(columns=['One', 'Gender']),
covar=data.drop(columns=['test', 'One', 'Gender']).columns
)[['X', 'Y', 'r']].set_index(['X', 'Y']).squeeze()
assert np.allclose(_pcorr.reindex(_pwcorr.index), _pwcorr)
# Partial pairwise with missing values
data.loc[[4, 5, 8, 20, 22], 'Age'] = np.nan
data.loc[[10, 12], 'Neuroticism'] = np.nan
pairwise_corr(data)
pairwise_corr(data, covar='Age')
# Listwise deletion
assert (pairwise_corr(data, covar='Age',
nan_policy='listwise')['n'].nunique() == 1)
assert pairwise_corr(data, nan_policy='listwise')['n'].nunique() == 1
######################################################################
# MultiIndex columns
from numpy.random import random as rdm
# Create MultiIndex dataframe
columns = pd.MultiIndex.from_tuples([('Behavior', 'Rating'),
('Behavior', 'RT'),
('Physio', 'BOLD'),
('Physio', 'HR'),
('Psycho', 'Anxiety')])
data = pd.DataFrame(dict(Rating=rdm(size=10),
RT=rdm(size=10),
BOLD=rdm(size=10),
HR=rdm(size=10),
Anxiety=rdm(size=10)))
data.columns = columns
pairwise_corr(data, method='spearman')
stats = pairwise_corr(data, columns=[('Behavior', 'Rating')])
assert stats.shape[0] == data.shape[1] - 1
pairwise_corr(data, columns=[('Behavior', 'Rating'),
('Behavior', 'RT')])
st1 = pairwise_corr(data, columns=[[('Behavior', 'Rating'),
('Behavior', 'RT')], None])
st2 = pairwise_corr(data, columns=[[('Behavior', 'Rating'),
('Behavior', 'RT')]])
assert st1['X'].equals(st2['X'])
st3 = pairwise_corr(data, columns=[[('Behavior', 'Rating')],
[('Behavior', 'RT'),
('Physio', 'BOLD')]])
assert st3.shape[0] == 2
# With covar
pairwise_corr(data, covar=[('Psycho', 'Anxiety')])
pairwise_corr(data, columns=[('Behavior', 'Rating')],
covar=[('Psycho', 'Anxiety')])
pairwise_corr(data, covar=data.columns)
# Test against pcorr:
_pcorr = data.pcorr()
_pcorr.index = _pcorr.index.tolist() # MultiIndex to tuples
_pcorr.columns = _pcorr.columns.tolist() # MultiIndex to tuples
_pwcorr = pairwise_corr(
data, covar=data.columns
)[['X', 'Y', 'r']].set_index(['X', 'Y']).squeeze()
assert np.allclose(_pcorr.stack().reindex(_pwcorr.index), _pwcorr)
# With missing values
data.iloc[2, [2, 3]] = np.nan
data.iloc[[1, 4], [1, 4]] = np.nan
assert pairwise_corr(data, nan_policy='listwise')['n'].nunique() == 1
assert pairwise_corr(data, nan_policy='pairwise')['n'].nunique() == 3
assert (pairwise_corr(data, columns=[('Behavior', 'Rating')],
covar=[('Psycho', 'Anxiety')],
nan_policy='listwise')['n'].nunique() == 1)
assert (pairwise_corr(data, columns=[('Behavior', 'Rating')],
covar=[('Psycho', 'Anxiety')],
nan_policy='pairwise')['n'].nunique() == 2)
