def test_pairwiseproptest(self):
ppt = smprop.proportions_chisquare_allpairs(self.n_success, self.nobs,
multitest_method=None)
assert_almost_equal(ppt.pvals_raw, self.res_ppt_pvals_raw)
ppt = smprop.proportions_chisquare_allpairs(self.n_success, self.nobs,
multitest_method='h')
assert_almost_equal(ppt.pval_corrected(), self.res_ppt_pvals_holm)
pptd = smprop.proportions_chisquare_pairscontrol(self.n_success,
self.nobs, multitest_method='hommel')
assert_almost_equal(pptd.pvals_raw, ppt.pvals_raw[:len(self.nobs) - 1],
decimal=13)
def test_pairwiseproptest(self):
ppt = smprop.proportions_chisquare_allpairs(self.n_success, self.nobs,
multitest_method=None)
assert_almost_equal(ppt.pvals_raw, self.res_ppt_pvals_raw)
ppt = smprop.proportions_chisquare_allpairs(self.n_success, self.nobs,
multitest_method='h')
assert_almost_equal(ppt.pval_corrected(), self.res_ppt_pvals_holm)
pptd = smprop.proportions_chisquare_pairscontrol(self.n_success,
self.nobs, multitest_method='hommel')
assert_almost_equal(pptd.pvals_raw, ppt.pvals_raw[:len(self.nobs) - 1],
decimal=13)
def test_number_pairs_1493(self):
ppt = smprop.proportions_chisquare_allpairs(self.n_success[:3],
self.nobs[:3],
multitest_method=None)
assert_equal(len(ppt.pvals_raw), 3)
idx = [0, 1, 3]
assert_almost_equal(ppt.pvals_raw, self.res_ppt_pvals_raw[idx])
def test_number_pairs_1493(self):
ppt = smprop.proportions_chisquare_allpairs(self.n_success[:3],
self.nobs[:3],
multitest_method=None)
assert_equal(len(ppt.pvals_raw), 3)
idx = [0, 1, 3]
assert_almost_equal(ppt.pvals_raw, self.res_ppt_pvals_raw[idx])
'N/A - No control group (comparing every group against all other variants)'
)
control_type = st.radio('Select Control Group', control_lst)
st.markdown(
'*__Example:__*<br /> _In our weekday example, if Monday is the control group, you want to compare email open rate for:_ <br /> - Monday vs Wednesday <br /> - Monday vs Friday <br />',
unsafe_allow_html=True)
if st.button('Calculate '):
if type(try_data_tbl(g_1, g_2, g_3)) == type(pd.DataFrame()):
st.write(try_data_tbl(g_1, g_2, g_3))
st.text(
'The calculation uses the holm-sidak method to adjust p-values in multiple tests.'
)
results = proportions_chisquare_allpairs(np.array(g_2),
np.array(g_3))
#st.write(results)
p_values = results.pval_corrected()
pair_names = results.all_pairs_names
sig = ''
nsig = ''
if control_type == 'N/A - No control group (comparing every group against all other variants)':
for i in range(len(p_values)):
t = f'__{g_1[int(pair_names[i][1])]}__ ({str(round(g_2[int(pair_names[i][1])]/g_3[int(pair_names[i][1])]*100,4))}%) and __{g_1[int(pair_names[i][4])]}__ ({str(round(g_2[int(pair_names[i][4])]/g_3[int(pair_names[i][4])]*100,4))}%) are'
if float(p_values[i]) < a:
sig = sig + (
'\n' + f'{t}' +
' <span style="color:green">significantly different</span> <br />'
)
elif float(p_values[i]) >= a:
nsig = nsig + (
ad_cats_table[ad_cats_table.isnull()] = 0
ad_cats_freq = ad_cats_table.apply(lambda x: x/sum(x), 1)
# no_rem is short for no remarketing, which is what happens when you're
# shown ads for a site you've visited.
ad_cats_table_no_rem = pd.DataFrame(all_ad_cats_no_remarketing).transpose()
ad_cats_table_no_rem[ad_cats_table_no_rem.isnull()] = 0
ad_cats_freq_no_rem = ad_cats_table_no_rem.apply(lambda x: x/sum(x), 1)
ad_cats_freq.to_csv('ad_cats_freq.tsv', sep='\t')
ad_cats_table.to_csv('ad_cats_counts.tsv', sep='\t')
ad_cats_table_no_rem.to_csv('ad_cats_counts_no_rem.tsv', sep='\t')
ad_cats_freq_no_rem.to_csv('ad_cats_freq_no_rem.tsv', sep='\t')
# do some statistical tests... chisquare?
chi2_cats = proportions_chisquare_allpairs(ad_cats_table, ad_cats_table.sum(1))
chi2_cats_no_rem = proportions_chisquare_allpairs(ad_cats_table_no_rem, ad_cats_table_no_rem.sum(1))
# correlation coefficient
ad_cats_corr = ad_cats_freq.T.corr()
ad_cats_corr_no_rem = ad_cats_freq_no_rem.T.corr()
# create bar plots for the ads targeted in each category?
xlab = np.arange(ad_cats_table.shape[1])
for category in ad_cats_table.index:
plt.clf()
plt.bar(xlab, ad_cats_freq.loc[category])
plt.ylim(0, 0.3)
plt.xticks(xlab+0.5, ad_cats_table.columns, rotation='vertical')
plt.title(category.split('_')[0])
plt.subplots_adjust()
