def test_neg_correlated_features():
trans_X_neg = (CBPM(corr_sign='neg')
.fit_transform(df_iris[X_neg], df_iris[y])
)
trans_X_pos_neg = (CBPM(corr_sign='neg')
.fit_transform(df_iris[X_posneg], df_iris[y])
)
trans_man = df_iris[X_neg].values.mean(axis=1)
assert_array_equal(trans_X_neg, trans_X_pos_neg)
assert_array_equal(trans_X_neg, trans_man)
def test_warn_posneg_no_feat():
with pytest.warns(RuntimeWarning,
match='No feature is significant'):
trans = (CBPM(corr_sign='posneg')
.fit_transform(df_shuffled_X[X_posneg], df_iris[y])
)
assert (trans == df_iris[y].values.mean()).all()
def test_posneg_correlated_features():
trans_X_posneg = (CBPM(corr_sign='posneg')
.fit_transform(df_iris[X_posneg], df_iris[y])
)
trans_man_pos = df_iris[X_pos].values.mean(axis=1)
trans_man_neg = df_iris[X_neg].values.mean(axis=1)
trans_man = np.concatenate([trans_man_pos.reshape(-1, 1),
trans_man_neg.reshape(-1, 1)], axis=1)
assert_array_equal(trans_X_posneg, trans_man)
def test_lower_sign_threshhold():
# I have checked before that only these 2 have pvalues under 1e-50
trans_posneg = (CBPM(corr_sign='pos', significance_threshold=1e-50)
.fit_transform(df_iris[X_pos], df_iris[y])
)
trans_man = df_iris[['petal_length', 'petal_width']].values.mean(axis=1)
assert_array_equal(trans_posneg, trans_man)
def test_warn_posneg_no_pos_feat():
with pytest.warns(RuntimeWarning,
match='No feature with significant positive'):
trans_posneg = (CBPM(corr_sign='posneg')
.fit_transform(df_iris[X_neg], df_iris[y])
)
trans_man = df_iris[X_neg].values.mean(axis=1)
assert_array_equal(trans_posneg, trans_man)
def test_lower_sign_threshhold_no_sig():
# I have checked before there are no under 1e-100
with pytest.warns(RuntimeWarning,
match='No feature is significant'):
trans_posneg = (CBPM(corr_sign='pos',
significance_threshold=1e-100)
.fit_transform(df_iris[X_pos], df_iris[y])
)
assert (trans_posneg == df_iris[y].values.mean()).all()
def test_neg_weighted_correlated_features():
trans_X_neg = (CBPM(corr_sign='neg', weight_by_corr=True)
.fit_transform(df_iris[X_neg], df_iris[y])
)
corr = df_iris[X_neg].apply(
lambda col: pearsonr(col, df_iris[y])[0]).values
trans_man = np.average(
df_iris[X_neg].values,
weights=corr,
axis=1)
assert_array_equal(trans_X_neg, trans_man)
def test_spearman():
# I have checked before all are still sign with spearman
trans_posneg = (CBPM(corr_method=spearmanr)
.fit_transform(df_iris[X_posneg], df_iris[y])
)
trans_man_pos = df_iris[X_pos].values.mean(axis=1)
trans_man_neg = df_iris[X_neg].values.mean(axis=1)
trans_man = np.concatenate([trans_man_pos.reshape(-1, 1),
trans_man_neg.reshape(-1, 1)], axis=1)
assert_array_equal(trans_posneg, trans_man)
def test_posneg_weighted_correlated_features():
trans_X_posneg = (CBPM(corr_sign='posneg', weight_by_corr=True)
.fit_transform(df_iris[X_posneg], df_iris[y])
)
corr_pos = df_iris[X_pos].apply(
lambda col: pearsonr(col, df_iris[y])[0]).values
corr_neg = df_iris[X_neg].apply(
lambda col: pearsonr(col, df_iris[y])[0]).values
trans_man_pos = np.average(
df_iris[X_pos].values,
weights=corr_pos,
axis=1)
trans_man_neg = np.average(
df_iris[X_neg].values,
weights=corr_neg,
axis=1)
trans_man = np.concatenate([trans_man_pos.reshape(-1, 1),
trans_man_neg.reshape(-1, 1)], axis=1)
assert_array_equal(trans_X_posneg, trans_man)