#!/usr/bin/env python3 import os import sys from statistics import variance import numpy as np import pandas as pd from scipy import stats CURDIR = os.path.dirname(os.path.realpath(__file__)) sys.path.insert(0, os.path.join(CURDIR, "helpers")) from pure_http_client import ClickHouseClient # unpooled variance z-test for means of two samples def twosample_mean_ztest(rvs1, rvs2, alpha=0.05): mean_rvs1 = np.mean(rvs1) mean_rvs2 = np.mean(rvs2) var_pop_rvs1 = variance(rvs1) var_pop_rvs2 = variance(rvs2) se = np.sqrt(var_pop_rvs1 / len(rvs1) + var_pop_rvs2 / len(rvs2)) z_stat = (mean_rvs1 - mean_rvs2) / se p_val = 2 * stats.norm.cdf(-1 * abs(z_stat)) z_a = stats.norm.ppf(1 - alpha / 2) ci_low = (mean_rvs1 - mean_rvs2) - z_a * se ci_high = (mean_rvs1 - mean_rvs2) + z_a * se return z_stat, p_val, ci_low, ci_high def test_and_check(name, a, b, t_stat, p_value, ci_low, ci_high, precision=1e-2): client = ClickHouseClient() client.query("DROP TABLE IF EXISTS ztest;") client.query("CREATE TABLE ztest (left Float64, right UInt8) ENGINE = Memory;") client.query( "INSERT INTO ztest VALUES {};".format( ", ".join(["({},{})".format(i, 0) for i in a]) ) ) client.query( "INSERT INTO ztest VALUES {};".format( ", ".join(["({},{})".format(j, 1) for j in b]) ) ) real = client.query_return_df( "SELECT roundBankers({}(left, right).1, 16) as t_stat, ".format(name) + "roundBankers({}(left, right).2, 16) as p_value, ".format(name) + "roundBankers({}(left, right).3, 16) as ci_low, ".format(name) + "roundBankers({}(left, right).4, 16) as ci_high ".format(name) + "FROM ztest FORMAT TabSeparatedWithNames;" ) real_t_stat = real["t_stat"][0] real_p_value = real["p_value"][0] real_ci_low = real["ci_low"][0] real_ci_high = real["ci_high"][0] assert ( abs(real_t_stat - np.float64(t_stat)) < precision ), "clickhouse_t_stat {}, py_t_stat {}".format(real_t_stat, t_stat) assert ( abs(real_p_value - np.float64(p_value)) < precision ), "clickhouse_p_value {}, py_p_value {}".format(real_p_value, p_value) assert ( abs(real_ci_low - np.float64(ci_low)) < precision ), "clickhouse_ci_low {}, py_ci_low {}".format(real_ci_low, ci_low) assert ( abs(real_ci_high - np.float64(ci_high)) < precision ), "clickhouse_ci_high {}, py_ci_high {}".format(real_ci_high, ci_high) client.query("DROP TABLE IF EXISTS ztest;") def test_mean_ztest(): rvs1 = np.round(stats.norm.rvs(loc=1, scale=5, size=500), 2) rvs2 = np.round(stats.norm.rvs(loc=10, scale=5, size=500), 2) s, p, cl, ch = twosample_mean_ztest(rvs1, rvs2) test_and_check( "meanZTest(%f, %f, 0.95)" % (variance(rvs1), variance(rvs2)), rvs1, rvs2, s, p, cl, ch, ) rvs1 = np.round(stats.norm.rvs(loc=0, scale=5, size=500), 2) rvs2 = np.round(stats.norm.rvs(loc=0, scale=5, size=500), 2) s, p, cl, ch = twosample_mean_ztest(rvs1, rvs2) test_and_check( "meanZTest(%f, %f, 0.95)" % (variance(rvs1), variance(rvs2)), rvs1, rvs2, s, p, cl, ch, ) rvs1 = np.round(stats.norm.rvs(loc=2, scale=10, size=512), 2) rvs2 = np.round(stats.norm.rvs(loc=5, scale=20, size=1024), 2) s, p, cl, ch = twosample_mean_ztest(rvs1, rvs2) test_and_check( "meanZTest(%f, %f, 0.95)" % (variance(rvs1), variance(rvs2)), rvs1, rvs2, s, p, cl, ch, ) rvs1 = np.round(stats.norm.rvs(loc=0, scale=10, size=1024), 2) rvs2 = np.round(stats.norm.rvs(loc=0, scale=10, size=512), 2) s, p, cl, ch = twosample_mean_ztest(rvs1, rvs2) test_and_check( "meanZTest(%f, %f, 0.95)" % (variance(rvs1), variance(rvs2)), rvs1, rvs2, s, p, cl, ch, ) if __name__ == "__main__": test_mean_ztest() print("Ok.")