def test_compare_branches():
df = pd.DataFrame(columns=['branch', 'val'],
index=range(1000),
dtype='float')
df.iloc[::2, 0] = 'control'
df.iloc[1::2, 0] = 'test'
df.iloc[:300, 1] = range(300)
with pytest.raises(ValueError):
mabssf.compare_branches(df, 'val', thresholds=[0., 15.9])
df.iloc[300:, 1] = 0
res = mabssf.compare_branches(df, 'val', thresholds=[0., 15.9])
assert res['individual']['control'].loc[0.,
'0.025'] == pytest.approx(st.beta(
149 + 1, 351 + 1).ppf(0.025),
abs=1e-6)
assert res['individual']['control'].loc[0.,
'0.5'] == pytest.approx(0.298537,
abs=1e-6)
assert res['individual']['test'].loc[15.9,
'0.025'] == pytest.approx(st.beta(
142 + 1, 358 + 1).ppf(0.025),
abs=1e-6)
assert res['individual']['test'].loc[15.9,
'0.5'] == pytest.approx(0.284575,
abs=1e-6)
def test_compare_branches():
df = pd.DataFrame(columns=["branch", "val"],
index=range(1000),
dtype="float")
df.iloc[::2, 0] = "control"
df.iloc[1::2, 0] = "test"
df.iloc[:300, 1] = range(300)
with pytest.raises(ValueError):
mabssf.compare_branches(df, "val", thresholds=[0.0, 15.9])
df.iloc[300:, 1] = 0
res = mabssf.compare_branches(df, "val", thresholds=[0.0, 15.9])
assert res["individual"]["control"].loc[0.0,
"0.025"] == pytest.approx(st.beta(
149 + 1, 351 + 1).ppf(0.025),
abs=1e-6)
assert res["individual"]["control"].loc[0.0,
"0.5"] == pytest.approx(0.298537,
abs=1e-6)
assert res["individual"]["test"].loc[15.9,
"0.025"] == pytest.approx(st.beta(
142 + 1, 358 + 1).ppf(0.025),
abs=1e-6)
assert res["individual"]["test"].loc[15.9,
"0.5"] == pytest.approx(0.284575,
abs=1e-6)
def test_few_auto_thresholds():
df = pd.DataFrame(columns=['branch', 'val'], index=range(1000))
df.iloc[::2, 0] = 'control'
df.iloc[1::2, 0] = 'test'
df.fillna(0, inplace=True)
df.iloc[20:30] = 1
df.iloc[405] = 100
res = mabssf.compare_branches(df, 'val')
assert set(res['individual']['test'].index) == {0, 1}
assert set(res['individual']['control'].index) == {0, 1}
def test_few_auto_thresholds():
df = pd.DataFrame(columns=["branch", "val"], index=range(1000))
df.iloc[::2, 0] = "control"
df.iloc[1::2, 0] = "test"
df.fillna(0, inplace=True)
df.iloc[20:30] = 1
df.iloc[405] = 100
res = mabssf.compare_branches(df, "val")
assert set(res["individual"]["test"].index) == {0, 1}
assert set(res["individual"]["control"].index) == {0, 1}
def plot_survival(df, col_label, ref_branch_label='control', thresholds=None):
data = masf.compare_branches(df, col_label, ref_branch_label, thresholds)
fig, (ax1, ax2) = plt.subplots(nrows=2, sharex=True, figsize=(6, 10))
for ax in (ax1, ax2):
ax.set_prop_cycle('color', plt.cm.tab20b(np.linspace(0, 1, 5)))
plot_means_line(ax1, data['individual'], ref_branch_label)
plot_uplifts_line(ax2, data['comparative'])
ax1.set_title(
'Survival fn: Fraction of users with {} > x'.format(col_label))
ax1.set_ylim((0, 1))
ax1.set_ylabel('Fraction of users')
ax1.legend()
ax2.set_xlabel(col_label)
ax2.set_ylabel('Uplift relative to {}'.format(ref_branch_label))
fig.tight_layout()
return fig