def test_confint(testdata):
result = confidence_interval(testdata, control_label='A')
c_means1 = CompareMeans(DescrStatsW(testdata['kpi1']['B']),
DescrStatsW(testdata['kpi1']['A']))
c_means2 = CompareMeans(DescrStatsW(testdata['kpi2']['B']),
DescrStatsW(testdata['kpi2']['A']))
expected1 = c_means1.tconfint_diff()
expected2 = c_means2.zconfint_diff()
assert result['B']['kpi1'] == expected1
assert result['B']['kpi2'] == expected2
def mean_diff_confint_ind(sample1, sample2, alpha=0.05):
"""Доверительный интервал разности средних для двух независимых выборок
Parameters
----------
sample1 : array_like
Первая выборка
sample2 : array_like
Вторая выборка
alpha : float in (0, 1)
Уровень доверия, рассчитывается как ``1-alpha``
Returns
-------
lower, upper : floats
Левая и правая граница доверительного интервала
"""
cm = CompareMeans(DescrStatsW(sample1), DescrStatsW(sample2))
return cm.tconfint_diff(alpha=alpha)
def test_ttest_2sample(self):
x1, x2 = self.x1, self.x2
x1r, x2r = self.x1r, self.x2r
w1, w2 = self.w1, self.w2
#Note: stats.ttest_ind handles 2d/nd arguments
res_sp = stats.ttest_ind(x1r, x2r)
assert_almost_equal(ttest_ind(x1, x2, weights=(w1, w2))[:2],
res_sp, 14)
#check correct ttest independent of user ddof
cm = CompareMeans(DescrStatsW(x1, weights=w1, ddof=0),
DescrStatsW(x2, weights=w2, ddof=1))
assert_almost_equal(cm.ttest_ind()[:2], res_sp, 14)
cm = CompareMeans(DescrStatsW(x1, weights=w1, ddof=1),
DescrStatsW(x2, weights=w2, ddof=2))
assert_almost_equal(cm.ttest_ind()[:2], res_sp, 14)
cm0 = CompareMeans(DescrStatsW(x1, weights=w1, ddof=0),
DescrStatsW(x2, weights=w2, ddof=0))
cm1 = CompareMeans(DescrStatsW(x1, weights=w1, ddof=0),
DescrStatsW(x2, weights=w2, ddof=1))
cm2 = CompareMeans(DescrStatsW(x1, weights=w1, ddof=1),
DescrStatsW(x2, weights=w2, ddof=2))
res0 = cm0.ttest_ind(usevar='unequal')
res1 = cm1.ttest_ind(usevar='unequal')
res2 = cm2.ttest_ind(usevar='unequal')
assert_almost_equal(res1, res0, 14)
assert_almost_equal(res2, res0, 14)
#check confint independent of user ddof
res0 = cm0.tconfint_diff(usevar='pooled')
res1 = cm1.tconfint_diff(usevar='pooled')
res2 = cm2.tconfint_diff(usevar='pooled')
assert_almost_equal(res1, res0, 14)
assert_almost_equal(res2, res0, 14)
res0 = cm0.tconfint_diff(usevar='unequal')
res1 = cm1.tconfint_diff(usevar='unequal')
res2 = cm2.tconfint_diff(usevar='unequal')
assert_almost_equal(res1, res0, 14)
assert_almost_equal(res2, res0, 14)
def test_ttest_2sample(self):
x1, x2 = self.x1, self.x2
x1r, x2r = self.x1r, self.x2r
w1, w2 = self.w1, self.w2
#Note: stats.ttest_ind handles 2d/nd arguments
res_sp = stats.ttest_ind(x1r, x2r)
assert_almost_equal(
ttest_ind(x1, x2, weights=(w1, w2))[:2], res_sp, 14)
#check correct ttest independent of user ddof
cm = CompareMeans(DescrStatsW(x1, weights=w1, ddof=0),
DescrStatsW(x2, weights=w2, ddof=1))
assert_almost_equal(cm.ttest_ind()[:2], res_sp, 14)
cm = CompareMeans(DescrStatsW(x1, weights=w1, ddof=1),
DescrStatsW(x2, weights=w2, ddof=2))
assert_almost_equal(cm.ttest_ind()[:2], res_sp, 14)
cm0 = CompareMeans(DescrStatsW(x1, weights=w1, ddof=0),
DescrStatsW(x2, weights=w2, ddof=0))
cm1 = CompareMeans(DescrStatsW(x1, weights=w1, ddof=0),
DescrStatsW(x2, weights=w2, ddof=1))
cm2 = CompareMeans(DescrStatsW(x1, weights=w1, ddof=1),
DescrStatsW(x2, weights=w2, ddof=2))
res0 = cm0.ttest_ind(usevar='unequal')
res1 = cm1.ttest_ind(usevar='unequal')
res2 = cm2.ttest_ind(usevar='unequal')
assert_almost_equal(res1, res0, 14)
assert_almost_equal(res2, res0, 14)
#check confint independent of user ddof
res0 = cm0.tconfint_diff(usevar='pooled')
res1 = cm1.tconfint_diff(usevar='pooled')
res2 = cm2.tconfint_diff(usevar='pooled')
assert_almost_equal(res1, res0, 14)
assert_almost_equal(res2, res0, 14)
res0 = cm0.tconfint_diff(usevar='unequal')
res1 = cm1.tconfint_diff(usevar='unequal')
res2 = cm2.tconfint_diff(usevar='unequal')
assert_almost_equal(res1, res0, 14)
assert_almost_equal(res2, res0, 14)
# 右片側検定なら'larger'
d = DescrStatsW(np.array(X) - np.array(Y)) # 対標本の場合
d.ttest_mean(alternative=alt)[1] # p値
#> 0.0006415571512322235
d.tconfint_mean(alpha=a, alternative=alt) # 信頼区間
#> (-3.9955246743198867, -1.3644753256801117)
c = CompareMeans(DescrStatsW(X), DescrStatsW(Y)) # 対標本でない場合
ve = 'pooled' # 等分散を仮定する（デフォルト）．仮定しないなら'unequal'．
c.ttest_ind(alternative=alt, usevar=ve)[1] # p値
#> 0.000978530937238609
c.tconfint_diff(alpha=a, alternative=alt, usevar=ve) # 信頼区間
#> (-4.170905570517185, -1.1890944294828283)
### 4.4.4 独立性の検定（カイ2乗検定）
import pandas as pd
my_url = ('https://raw.githubusercontent.com/taroyabuki'
'/fromzero/master/data/smoker.csv')
my_data = pd.read_csv(my_url)
my_data.head()
#> alive smoker
#> 0 Yes No
#> 1 Yes No
#> 2 Yes No
#> 3 Yes No
def fn(control, test):
c_means = CompareMeans(DescrStatsW(test), DescrStatsW(control))
if _is_proportion(control, test):
return c_means.zconfint_diff()
else:
return c_means.tconfint_diff()
def mean_diff_confint_ind(sample1, sample2):
cm = CompareMeans(DescrStatsW(sample1), DescrStatsW(sample2))
return cm.tconfint_diff()
