def test_thresholds(fico):
threshold_data = threshold.find_thresholds(fico['rocs'],
fico['proportions'],
fico['base_rate'],
fico['base_rates'], COST_MATRIX)
assert_deep_almost_equal(FICO_THRESHOLD_DATA, threshold_data, **FICO_TOL)
def test_independence_thresholds(fico):
assert_deep_almost_equal(
FICO_INDEPENDENCE,
threshold.find_independence_thresholds(fico['rocs'],
fico['base_rates'],
fico['proportions'],
COST_MATRIX), **FICO_TOL)
def test_most_similar(w2v_small):
POSITIVE, NEGATIVE = ('doctor', 'she'), ('he', )
responsibly_results = most_similar(w2v_small, POSITIVE, NEGATIVE, topn=10)
gensim_results = w2v_small.most_similar(POSITIVE, NEGATIVE, topn=9)
assert responsibly_results[0][0] == 'doctor'
assert_deep_almost_equal(responsibly_results[1:],
gensim_results,
atol=0.01)
def test_generate_analogies(gender_biased_w2v_small):
"""Test generate_analogies method in GenderBiasWE.
Based on:
https://github.com/tolga-b/debiaswe/blob/master/tutorial_example1.ipynb
"""
analogies_df = (gender_biased_w2v_small.generate_analogies(
500, unrestricted=False))
analogies_df = analogies_df[['she', 'he']]
assert_deep_almost_equal(analogies_df.values, TOLGA_GENDER_ANALOGIES)
def test_calc_all_weat_defaults(w2v_small):
weat_5_results_default = (calc_all_weat(w2v_small,
(5, )).iloc[0].to_dict())
weat_5_results = (calc_all_weat(w2v_small, (5, ),
filter_by='model',
with_pvalue=True,
pvalue_kwargs={
'method': 'exact'
}).iloc[0].to_dict())
assert_deep_almost_equal(weat_5_results_default, weat_5_results)
def test_calc_weat_pleasant_attribute(w2v_small):
# pylint: disable=line-too-long
pvalue_kwargs = {'method': 'approximate'}
result_v1 = calc_weat_pleasant_unpleasant_attribute(
w2v_small,
WEAT_DATA[1]['first_target'],
WEAT_DATA[1]['second_target'],
pvalue_kwargs=pvalue_kwargs)
result_v1['p'] = round(result_v1['p'], 4)
result_v1['d'] = round(result_v1['d'], 4)
result_v1['s'] = round(result_v1['s'], 4)
result_v2 = (calc_all_weat(w2v_small, (1, ),
pvalue_kwargs=pvalue_kwargs).iloc[0].to_dict())
assert_deep_almost_equal(result_v1, result_v2, atol=0.01)
def test_calc_all_weat_index(w2v_small):
all_weat = calc_all_weat(w2v_small,
filter_by='model',
with_original_finding=True,
with_pvalue=True,
pvalue_kwargs={'method': 'approximate'})
for index in range(len(WEAT_DATA)):
single_weat = calc_all_weat(w2v_small,
weat_data=index,
filter_by='model',
with_original_finding=True,
with_pvalue=True,
pvalue_kwargs={'method': 'approximate'})
assert_deep_almost_equal(single_weat.iloc[0].to_dict(),
all_weat.iloc[index].to_dict(),
atol=0.01)
def test_separation_thresholds(fico):
assert_deep_almost_equal(
FICO_SEPARATION,
threshold.find_separation_thresholds(fico['rocs'], fico['base_rate'],
COST_MATRIX), **FICO_TOL)
def test_fnr_thresholds(fico):
assert_deep_almost_equal(
FICO_FNR,
threshold.find_fnr_thresholds(fico['rocs'], fico['base_rates'],
fico['proportions'], COST_MATRIX),
**FICO_TOL)
def test_min_cost_threshold(fico):
assert_deep_almost_equal(
FICO_MIN_COST,
threshold.find_min_cost_thresholds(fico['rocs'], fico['base_rates'],
fico['proportions'], COST_MATRIX),
**FICO_TOL)
def test_single_threshold(fico):
assert_deep_almost_equal(
FICO_SINGLE,
threshold.find_single_threshold(fico['rocs'], fico['base_rates'],
fico['proportions'], COST_MATRIX),
**FICO_TOL)
