def get_sample_size(self, beta=0.1):
"""
Calculate required sample size per group to obtain provided beta.
Parameters
----------
beta : float
Type 2 error rate.
Returns
-------
n : int
sample size per group.
"""
es = sms.proportion_effectsize(self.prop_null, self.prop_alt)
if self.alt_hypothesis == 'two_tailed':
n = tt_ind_solve_power(
effect_size=es,
alpha=self.alpha,
power=1 - beta,
alternative='two-sided',
)
else:
n = tt_ind_solve_power(
effect_size=es,
alpha=self.alpha,
power=1 - beta,
alternative='smaller')
return int(np.round(n, 0))
def _design_experiment(self, *args):
test_type, confidence_level, power, before_eff, after_eff = args
self.test_type = test_type
self.confidence_level = confidence_level
self.alpha = 1 - confidence_level
self.beta = power
self.effect_size = sms.proportion_effectsize(before_eff, after_eff)
def two_proportions_sample_size(p1, p2, alpha=0.05, power=0.8, frac=0.5):
ratio = frac / (1. - frac)
es = sms.proportion_effectsize(p1, p2)
n = np.floor(sms.NormalIndPower().solve_power(es,
power=power,
alpha=alpha,
ratio=ratio))
n1, n2 = n * ratio, n
return n1, n2
def post(self):
sig_level = 0.05
power = 0.8
body = request.get_json(silent=True) or {}
p1 = body['p1']
p2 = body['p2']
p1_and_p2 = sms.proportion_effectsize(p1, p2)
sample_size = sms.NormalIndPower().solve_power(p1_and_p2,
power=power,
alpha=sig_level)
return {
'error': False,
'message': 'The required sample size for this campaign is:',
'data': round(sample_size),
'status_code': 200
}, 200
def proportions_two_samplesize(p1, p2, frac=0.5, power=0.8, alpha=0.05):
"""Размер выборки для двух долей
Parameters
----------
p1 : float in (0, 1)
Улучшаемый показатель, например, 0.1
p2 : float in (0, 1)
Требуемое значение показателя, например, 0.1 * 1.2 (на 20 процентов больше p1)
frac : float in (0, 1)
Пропорция контрольной и общего размера теста, например, 0.2 - это 20% от всего эксперимента
power : float in (0, 1)
Мощность, по умолчанию 0.8
alpha : float in (0, 1)
Достигаемый уровень значимсоти, по-умолчанию 0.05
Returns
-------
n1, n2 : float
Необходимое количество наблюдений в контрольной и тестовой группах,
сумма показывает общее количество необходимых наблюдений
Notes
-----
Используется для вычисления размера требуемой выборки при проведении AB-теста
p1 - показатель который необходимо улучшить до уровня p2.
В примере:
p1 - 0.1 (10%)
p2 - 0.1*1.2 - требуется улучшить на 20%
frac - 0.2 - 20% контроль, 80% тест
Examples
--------
>>> proportions_two_samplesize(0.1, 0.1 * 1.2, frac=0.2)
(2396.5, 9586.0)
"""
ratio = frac / (1. - frac)
es = proportion_effectsize(p1, p2)
n = np.floor(NormalIndPower().solve_power(es,
power=power,
alpha=alpha,
ratio=ratio))
n1, n2 = n * ratio, n
return n1, n2
def get_sample_size_for_binomial(p0, p1, power = 0.8, significance = 0.05):
"""
Calculate the sample size for the binomial distribution
Parameters
----------
p0 : [0, 1] current proportion value
p1 : [0, 1] expected proportion value
power : power of the test, e.g. 0.8,
is one minus the probability of a type II error.
Power is the probability that the test correctly
rejects the Null Hypothesis if the Alternative Hypothesis is true.
significance : significance level, e.g. 0.05, is the probability
of a type I error, that is wrong rejections
if the Null Hypothesis is true.
Returns
-------
required_n : Number of observations per sample.
"""
effect_size = sms.proportion_effectsize(p0, p1)
required_n = sms.NormalIndPower().solve_power(effect_size, power=power, alpha=significance)
return ceil(required_n)
df = pd.read_csv('ab_data.csv')
df.head()
pd.crosstab(index=df['group'], columns=df['landing_page'])
df.groupby('group')['converted'].mean()
df_sub = df.drop_duplicates(subset='user_id', keep="first")
print("Number of rows of data:", len(df))
print("Number of rows of data after removing multiple users:", len(df_sub))
df_sub.groupby('group')['converted'].mean()
# Get least number of sample size
effect = sms.proportion_effectsize(0.12, 0.14)
alpha = 0.05
power = 0.8
analysis = TTestIndPower()
result = analysis.solve_power(effect,
power=power,
nobs1=None,
ratio=1.0,
alpha=alpha)
print('Sample Size: %.3f' % result)
import math
n = math.ceil(result)
#Sample for n control and n treatment group
df_control = df_sub.loc[df_sub['group'] == 'control']
# Net Conversion - number of payments divided by number of clicks
payments_cont = control["Payments"].sum()
payments_exp = experiment["Payments"].sum()
NC_cont = payments_cont / clicks_cont
NC_exp = payments_exp / clicks_exp
NC_pooled = (payments_cont + payments_exp) / (clicks_cont + clicks_exp)
NC_sd_pooled = mt.sqrt(NC_pooled * (1 - NC_pooled) *
(1 / clicks_cont + 1 / clicks_exp))
NC_ME = round(get_z_score(1 - alpha / 2) * NC_sd_pooled, 4)
NC_diff = round(NC_exp - NC_cont, 4)
# print("The change due to the experiment is",NC_diff*100,"%")
# print("Confidence Interval: [",NC_diff-NC_ME,",",NC_diff+NC_ME,"]")
# print ("The change is statistically significant if the CI doesn't include 0. In that case, it is practically significant if",NC["d_min"],"is not in the CI as well.")
# Case91_嘗試以函數算出樣本數_Calculating effect size based on our expected rates
effect_size = sms.proportion_effectsize(GC["p"] - 1.0 * GC["d_min"],
GC["p"] + 0.0 * GC["d_min"])
required_n = sms.NormalIndPower().solve_power(effect_size,
power=0.8,
alpha=0.05,
ratio=1)
required_n = ceil(required_n)
print(effect_size, required_n)
# Case02-自行開發雙樣本比例的信賴區間函數
def two_proprotions_confint(success_a,
size_a,
success_b,
size_b,
significance=0.05):
prop_a = success_a / size_a
import pandas as pd
import scipy.stats as stats
import statsmodels.stats.api as sms
import matplotlib as mpl
import matplotlib.pyplot as plt
import seaborn as sns
from math import ceil
# Some plot styling preferences
plt.style.use('seaborn-whitegrid')
font = {'family': 'Helvetica', 'weight': 'bold', 'size': 14}
mpl.rc('font', **font)
# Calculating effect size based on our expected rates
effect_size = sms.proportion_effectsize(0.13, 0.15)
required_n = sms.NormalIndPower().solve_power(effect_size,
power=0.8,
alpha=0.05,
ratio=1)
# Calculating sample size needed
required_n = ceil(required_n)
# Rounding up to next whole number
# print(required_n)
#展示實驗資料
df = pd.read_csv('ab_data.csv')
# To make sure all the control group are seeing the old page and viceversa
# 用 crosstab 將 landing_page 當作 column，group 當作 row
# df=pd.crosstab(df['group'], df['landing_page'])
import matplotlib.pyplot as plt
import seaborn as sns
from math import ceil
%matplotlib inline
# Some plot styling preferences
plt.style.use('seaborn-whitegrid')
font = {'family' : 'Helvetica',
'weight' : 'bold',
'size' : 14}
mpl.rc('font', **font)
# calculate effect size by propotion
effect_size = sms.proportion_effectsize(0.13, 0.15) # Calculating effect size based on our expected rates
required_n = sms.NormalIndPower().solve_power(
effect_size,
power=0.8,
alpha=0.05,
ratio=1
) # Calculating sample size needed
required_n = ceil(required_n) # Rounding up to next whole number
print(required_n)
# get dataframe info
df.info()