def get_peruser_diff_zstat(df,
nobs_name,
mean_name,
std_name,
test_group_column='test_group',
test='TEST',
control='CONTROL',
alpha=0.05,
alternative='two-sided'):
'''confidence interval for the difference in means.
Similar to https://www.statsmodels.org/dev/generated/statsmodels.stats.weightstats.CompareMeans.zconfint_diff.html#statsmodels.stats.weightstats.CompareMeans.zconfint_diff
Example: pre = s1.build_and_run_sql(grouping=['test_group','user_id'], sql_template=TEMPLATE_USER_STATS)
get_user_agg_zstat(pre, 'units','buyers mean', 'buyers std')
Parameters
----------
alpha : float
significance level for the confidence interval, coverage is
``1-alpha``
alternative : string
This specifies the alternative hypothesis for the test that
corresponds to the confidence interval.
The alternative hypothesis, H1, has to be one of the following :
'two-sided': H1: difference in means not equal to value (default)
'larger' : H1: difference in means larger than value
'smaller' : H1: difference in means smaller than value
Returns
-------
diff, zstat, pvalue, confint : floats
'''
summary = df.set_index(test_group_column)#.T.loc[[nobs_name, mean_name, std_name],[test, control]].to_dict()
d1 = summary.loc[test, :]
d2 = summary.loc[control,:]
diff = d1[mean_name] - d2[mean_name]
std_diff = np.sqrt(d1[std_name]**2/(d1[nobs_name]-1) + d2[std_name]**2/(d2[nobs_name]-1)) #Assume unequal variance
confint = wstats._zconfint_generic(diff, std_diff, alpha=alpha, alternative=alternative)
zstat, pvalue = wstats._zstat_generic2(diff, std_diff, alternative)
return d1[mean_name], d2[mean_name], diff/d2[mean_name], confint/d2[mean_name], zstat, pvalue
def proportions_ztest(count, nobs, value=None, alternative='two-sided',
prop_var=False):
'''test for proportions based on normal (z) test
Parameters
----------
count : integer or array_like
the number of successes in nobs trials. If this is array_like, then
the assumption is that this represents the number of successes for
each independent sample
nobs : integer
the number of trials or observations, with the same length as
count.
value : None or float or array_like
This is the value of the null hypothesis equal to the proportion in the
case of a one sample test. In the case of a two-sample test, the
null hypothesis is that prop[0] - prop[1] = value, where prop is the
proportion in the two samples
alternative : string in ['two-sided', 'smaller', 'larger']
The alternative hypothesis can be either two-sided or one of the one-
sided tests, smaller means that the alternative hypothesis is
``prop < value` and larger means ``prop > value``, or the corresponding
inequality for the two sample test.
prop_var : False or float in (0, 1)
If prop_var is false, then the variance of the proportion estimate is
calculated based on the sample proportion. Alternatively, a proportion
can be specified to calculate this variance. Common use case is to
use the proportion under the Null hypothesis to specify the variance
of the proportion estimate.
TODO: change options similar to propotion_ztost ?
Returns
-------
zstat : float
test statistic for the z-test
p-value : float
p-value for the z-test
Notes
-----
This uses a simple normal test for proportions. It should be the same as
running the mean z-test on the data encoded 1 for event and 0 for no event,
so that the sum corresponds to count.
In the one and two sample cases with two-sided alternative, this test
produces the same p-value as ``proportions_chisquare``, since the
chisquare is the distribution of the square of a standard normal
distribution.
(TODO: verify that this really holds)
TODO: add continuity correction or other improvements for small samples.
'''
prop = count * 1. / nobs
k_sample = np.size(prop)
if k_sample == 1:
diff = prop - value
elif k_sample == 2:
diff = prop[0] - prop[1] - value
else:
msg = 'more than two samples are not implemented yet'
raise NotImplementedError(msg)
p_pooled = np.sum(count) * 1. / np.sum(nobs)
nobs_fact = np.sum(1. / nobs)
if prop_var:
p_pooled = prop_var
var_ = p_pooled * (1 - p_pooled) * nobs_fact
std_diff = np.sqrt(var_)
from statsmodels.stats.weightstats import _zstat_generic2
return _zstat_generic2(diff, std_diff, alternative)
def test_poisson_2indep(count1,
exposure1,
count2,
exposure2,
ratio_null=1,
method='score',
alternative='two-sided',
etest_kwds=None):
'''test for ratio of two sample Poisson intensities
If the two Poisson rates are g1 and g2, then the Null hypothesis is
- H0: g1 / g2 = ratio_null
against one of the following alternatives
- H1_2-sided: g1 / g2 != ratio_null
- H1_larger: g1 / g2 > ratio_null
- H1_smaller: g1 / g2 < ratio_null
Parameters
----------
count1 : int
Number of events in first sample.
exposure1 : float
Total exposure (time * subjects) in first sample.
count2 : int
Number of events in second sample.
exposure2 : float
Total exposure (time * subjects) in second sample.
ratio: float
ratio of the two Poisson rates under the Null hypothesis. Default is 1.
method : string
Method for the test statistic and the p-value. Defaults to `'score'`.
Current Methods are based on Gu et. al 2008.
Implemented are 'wald', 'score' and 'sqrt' based asymptotic normal
distribution, and the exact conditional test 'exact-cond', and its
mid-point version 'cond-midp'. method='etest' and method='etest-wald'
provide pvalues from `etest_poisson_2indep` using score or wald
statistic respectively.
see Notes.
alternative : string
The alternative hypothesis, H1, has to be one of the following
- 'two-sided': H1: ratio of rates is not equal to ratio_null (default)
- 'larger' : H1: ratio of rates is larger than ratio_null
- 'smaller' : H1: ratio of rates is smaller than ratio_null
etest_kwds: dictionary
Additional parameters to be passed to the etest_poisson_2indep
function, namely ygrid.
Returns
-------
results : instance of HolderTuple class
The two main attributes are test statistic `statistic` and p-value
`pvalue`.
Notes
-----
- 'wald': method W1A, wald test, variance based on separate estimates
- 'score': method W2A, score test, variance based on estimate under Null
- 'wald-log': W3A
- 'score-log' W4A
- 'sqrt': W5A, based on variance stabilizing square root transformation
- 'exact-cond': exact conditional test based on binomial distribution
- 'cond-midp': midpoint-pvalue of exact conditional test
- 'etest': etest with score test statistic
- 'etest-wald': etest with wald test statistic
References
----------
Gu, Ng, Tang, Schucany 2008: Testing the Ratio of Two Poisson Rates,
Biometrical Journal 50 (2008) 2, 2008
See Also
--------
tost_poisson_2indep
etest_poisson_2indep
'''
# shortcut names
y1, n1, y2, n2 = count1, exposure1, count2, exposure2
d = n2 / n1
r = ratio_null
r_d = r / d
if method in ['score']:
stat = (y1 - y2 * r_d) / np.sqrt((y1 + y2) * r_d)
dist = 'normal'
elif method in ['wald']:
stat = (y1 - y2 * r_d) / np.sqrt(y1 + y2 * r_d**2)
dist = 'normal'
elif method in ['sqrt']:
stat = 2 * (np.sqrt(y1 + 3 / 8.) - np.sqrt((y2 + 3 / 8.) * r_d))
stat /= np.sqrt(1 + r_d)
dist = 'normal'
elif method in ['exact-cond', 'cond-midp']:
from statsmodels.stats import proportion
bp = r_d / (1 + r_d)
y_total = y1 + y2
stat = None
# TODO: why y2 in here and not y1, check definition of H1 "larger"
pvalue = proportion.binom_test(y1,
y_total,
prop=bp,
alternative=alternative)
if method in ['cond-midp']:
# not inplace in case we still want binom pvalue
pvalue = pvalue - 0.5 * stats.binom.pmf(y1, y_total, bp)
dist = 'binomial'
elif method.startswith('etest'):
if method.endswith('wald'):
method_etest = 'wald'
else:
method_etest = 'score'
if etest_kwds is None:
etest_kwds = {}
stat, pvalue = etest_poisson_2indep(count1,
exposure1,
count2,
exposure2,
ratio_null=ratio_null,
method=method_etest,
alternative=alternative,
**etest_kwds)
dist = 'poisson'
else:
raise ValueError('method not recognized')
if dist == 'normal':
stat, pvalue = _zstat_generic2(stat, 1, alternative)
rates = (y1 / n1, y2 / n2)
ratio = rates[0] / rates[1]
res = HolderTuple(statistic=stat,
pvalue=pvalue,
distribution=dist,
method=method,
alternative=alternative,
rates=rates,
ratio=ratio,
ratio_null=ratio_null)
return res
def proportions_ztest(count, nobs, value=None, alternative='two-sided',
prop_var=False):
"""
Test for proportions based on normal (z) test
Parameters
----------
count : integer or array_like
the number of successes in nobs trials. If this is array_like, then
the assumption is that this represents the number of successes for
each independent sample
nobs : integer or array-like
the number of trials or observations, with the same length as
count.
value : float, array_like or None, optional
This is the value of the null hypothesis equal to the proportion in the
case of a one sample test. In the case of a two-sample test, the
null hypothesis is that prop[0] - prop[1] = value, where prop is the
proportion in the two samples. If not provided value = 0 and the null
is prop[0] = prop[1]
alternative : string in ['two-sided', 'smaller', 'larger']
The alternative hypothesis can be either two-sided or one of the one-
sided tests, smaller means that the alternative hypothesis is
``prop < value` and larger means ``prop > value``, or the corresponding
inequality for the two sample test.
prop_var : False or float in (0, 1)
If prop_var is false, then the variance of the proportion estimate is
calculated based on the sample proportion. Alternatively, a proportion
can be specified to calculate this variance. Common use case is to
use the proportion under the Null hypothesis to specify the variance
of the proportion estimate.
Returns
-------
zstat : float
test statistic for the z-test
p-value : float
p-value for the z-test
Examples
--------
>>> count = 5
>>> nobs = 83
>>> value = .05
>>> stat, pval = proportions_ztest(count, nobs, value)
>>> print('{0:0.3f}'.format(pval))
0.695
>>> import numpy as np
>>> from statsmodels.stats.proportion import proportions_ztest
>>> count = np.array([5, 12])
>>> nobs = np.array([83, 99])
>>> stat, pval = proportions_ztest(counts, nobs)
>>> print('{0:0.3f}'.format(pval))
0.159
Notes
-----
This uses a simple normal test for proportions. It should be the same as
running the mean z-test on the data encoded 1 for event and 0 for no event
so that the sum corresponds to the count.
In the one and two sample cases with two-sided alternative, this test
produces the same p-value as ``proportions_chisquare``, since the
chisquare is the distribution of the square of a standard normal
distribution.
"""
# TODO: verify that this really holds
# TODO: add continuity correction or other improvements for small samples
# TODO: change options similar to propotion_ztost ?
count = np.asarray(count)
nobs = np.asarray(nobs)
if nobs.size == 1:
nobs = nobs * np.ones_like(count)
prop = count * 1. / nobs
k_sample = np.size(prop)
if value is None:
if k_sample == 1:
raise ValueError('value must be provided for a 1-sample test')
value = 0
if k_sample == 1:
diff = prop - value
elif k_sample == 2:
diff = prop[0] - prop[1] - value
else:
msg = 'more than two samples are not implemented yet'
raise NotImplementedError(msg)
p_pooled = np.sum(count) * 1. / np.sum(nobs)
nobs_fact = np.sum(1. / nobs)
if prop_var:
p_pooled = prop_var
var_ = p_pooled * (1 - p_pooled) * nobs_fact
std_diff = np.sqrt(var_)
from statsmodels.stats.weightstats import _zstat_generic2
return _zstat_generic2(diff, std_diff, alternative)
def proportions_ztest(count,
nobs,
value=None,
alternative='two-sided',
prop_var=False):
'''test for proportions based on normal (z) test
Parameters
----------
count : integer or array_like
the number of successes in nobs trials. If this is array_like, then
the assumption is that this represents the number of successes for
each independent sample
nobs : integer
the number of trials or observations, with the same length as
count.
value : None or float or array_like
This is the value of the null hypothesis equal to the proportion in the
case of a one sample test. In the case of a two-sample test, the
null hypothesis is that prop[0] - prop[1] = value, where prop is the
proportion in the two samples
alternative : string in ['two-sided', 'smaller', 'larger']
The alternative hypothesis can be either two-sided or one of the one-
sided tests, smaller means that the alternative hypothesis is
``prop < value` and larger means ``prop > value``, or the corresponding
inequality for the two sample test.
prop_var : False or float in (0, 1)
If prop_var is false, then the variance of the proportion estimate is
calculated based on the sample proportion. Alternatively, a proportion
can be specified to calculate this variance. Common use case is to
use the proportion under the Null hypothesis to specify the variance
of the proportion estimate.
TODO: change options similar to propotion_ztost ?
Returns
-------
zstat : float
test statistic for the z-test
p-value : float
p-value for the z-test
Notes
-----
This uses a simple normal test for proportions. It should be the same as
running the mean z-test on the data encoded 1 for event and 0 for no event,
so that the sum corresponds to count.
In the one and two sample cases with two-sided alternative, this test
produces the same p-value as ``proportions_chisquare``, since the
chisquare is the distribution of the square of a standard normal
distribution.
(TODO: verify that this really holds)
TODO: add continuity correction or other improvements for small samples.
'''
prop = count * 1. / nobs
k_sample = np.size(prop)
if k_sample == 1:
diff = prop - value
elif k_sample == 2:
diff = prop[0] - prop[1] - value
else:
msg = 'more than two samples are not implemented yet'
raise NotImplementedError(msg)
p_pooled = np.sum(count) * 1. / np.sum(nobs)
nobs_fact = np.sum(1. / nobs)
if prop_var:
p_pooled = prop_var
var_ = p_pooled * (1 - p_pooled) * nobs_fact
std_diff = np.sqrt(var_)
from statsmodels.stats.weightstats import _zstat_generic2
return _zstat_generic2(diff, std_diff, alternative)
for Id in IdLs:
BranchScoreDf.loc[Indx, Id] = abs(row[f"{Id}_Subs"] -
AvgLen[f"{Id}_Subs"]) #**2
BranchScoreDf['TreeFull'] = BranchScoreDf[IdLs].sum(axis=1)
BranchScoreDf.to_csv('BranchScoreDf.tsv',
sep='\t',
header=True,
index=True,
index_label='Id')
BraScStatSigDf = pd.DataFrame(dtype=np.float64)
ColLs = list(BranchScoreDf.columns)
for Indx, row in BranchScoreDf.iterrows():
for Col in ColLs:
BraScStatSigDf.at[Indx, Col] = weightstats._zstat_generic2(
row[Col],
BsDf.at[Indx, f"SE_{Col}"],
alternative='two-sided',
)[-1] / 2
BraScStatSigDf.to_csv('BraScStatSigDf.tsv',
sep='\t',
header=True,
index=True,
index_label='Id')
FdrCorrection(BraScStatSigDf)
BraScStatSigDf.to_csv('FdrBraScStatSigDf.tsv',
sep='\t',
header=True,
index=True,
index_label='Id')