{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"name": "ab-tests-with-python_作業解答.ipynb",
"provenance": []
},
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.6.3"
}
},
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "RTnv32wkA_Ox"
},
"source": [
"# **作業說明**\n",
"# (這是Udacity關於A/B Test的期末專題)\n",
"\n",
"Udacity希望了解，在免費14天試學網頁上，除了要信用卡資訊外，還想了解學生願意花多少小時學。如果少於某門檻(5小時)，就建議學生不要註冊，免費聽聽影音就好，免得浪費資源，降低學習成功率。\n",
"\n",
"我們的題目是，增加這個頁面，是否對Gross Conversion(GC)和Net Conversion (NC)在統計學上(Alpha=0.05，Power=0.8)有幫助(d=0.01/0.0075)，亦即統計上的顯著(Significant)。\n",
"\n",
"CI = click 數目\n",
"\n",
"GC = 註冊數/CI (聽了建議仍然註冊的比例)\n",
"\n",
"NC = 繳費數/CI (14天之後繳費且繼續的比例)\n",
"\n",
"我們期待GC比原來下降，但NC不降，這表示省去資源但收入不降。\n",
"\n",
"檔名：ab-tests-with-python.ipynb\n",
"\n",
"**作業目標**\n",
"\n",
"1. 經由範例程式，學習A/B Test 的步驟\n",
"2. 最低樣本數的計算方法\n",
"3. 自行開發信賴區間計算函數\n",
"\n",
"\n",
"\n",
"\n",
"\n",
"\n",
"\n",
"\n",
"\n",
"\n",
"\n"
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"background_save": true
},
"id": "QfCMcrfTFyMx"
},
"source": [
"#載入程式庫\n",
"import math as mt\n",
"import numpy as np\n",
"import pandas as pd\n",
"from scipy.stats import norm"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "_iZnYjxIFyMy"
},
"source": [
"#將基礎數據放入字典\n",
"baseline = {\"Cookies\":40000,\"Clicks\":3200,\"Enrollments\":660,\"CTP\":0.08,\"GConversion\":0.20625,\n",
" \"Retention\":0.53,\"NConversion\":0.109313}"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "rE-idI4vFyMy"
},
"source": [
"#調整大小到以Cookie為基準\n",
"baseline[\"Cookies\"] = 5000\n",
"baseline[\"Clicks\"]=baseline[\"Clicks\"]*(5000/40000)\n",
"baseline[\"Enrollments\"]=baseline[\"Enrollments\"]*(5000/40000)\n",
"baseline"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "gNpNShHKFyMz",
"outputId": "1c00edec-27d3-4729-88e8-ce58eeef0c99"
},
"source": [
"# 算出 Gross Conversion (GC) 的 p 和 n\n",
"# 還有 Stansard Deviation(sd) rounded to 4 decimal digits.\n",
"GC={}\n",
"GC[\"d_min\"]=0.01\n",
"GC[\"p\"]=baseline[\"GConversion\"]\n",
"#p is given in this case - or we could calculate it from enrollments/clicks\n",
"GC[\"n\"]=baseline[\"Clicks\"]\n",
"GC[\"sd\"]=round(mt.sqrt((GC[\"p\"]*(1-GC[\"p\"]))/GC[\"n\"]),4)\n",
"GC[\"sd\"]"
],
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"0.0202"
]
},
"metadata": {
"tags": []
},
"execution_count": 9
}
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "w_WjAIKkFyMz",
"outputId": "c924bda0-918c-4073-8e55-7cc3d67ad4e9"
},
"source": [
"# Retention(R) \n",
"\n",
"R={}\n",
"R[\"d_min\"]=0.01\n",
"R[\"p\"]=baseline[\"Retention\"]\n",
"R[\"n\"]=baseline[\"Enrollments\"]\n",
"R[\"sd\"]=round(mt.sqrt((R[\"p\"]*(1-R[\"p\"]))/R[\"n\"]),4)\n",
"R[\"sd\"]"
],
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"0.0549"
]
},
"metadata": {
"tags": []
},
"execution_count": 4
}
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "kwx3Of06FyMz",
"outputId": "efc60cde-6e3b-4bed-e5f6-a3b697b3d393"
},
"source": [
"# Net Conversion (NC)\n",
"NC={}\n",
"NC[\"d_min\"]=0.0075\n",
"NC[\"p\"]=baseline[\"NConversion\"]\n",
"NC[\"n\"]=baseline[\"Clicks\"]\n",
"NC[\"sd\"]=round(mt.sqrt((NC[\"p\"]*(1-NC[\"p\"]))/NC[\"n\"]),4)\n",
"NC[\"sd\"]"
],
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"0.0156"
]
},
"metadata": {
"tags": []
},
"execution_count": 5
}
]
},
{
"cell_type": "code",
"metadata": {
"id": "pBk7b5uMFyM0"
},
"source": [
"def get_sds(p,d):\n",
" sd1=mt.sqrt(2*p*(1-p))\n",
" sd2=mt.sqrt(p*(1-p)+(p+d)*(1-(p+d)))\n",
" x=[sd1,sd2]\n",
" return x"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "mB2im4rcFyM0"
},
"source": [
"#計算 Z-score\n",
"def get_z_score(alpha):\n",
" return norm.ppf(alpha)\n",
"\n",
"# 得到兩個(A/B)標準差\n",
"def get_sds(p,d):\n",
" sd1=mt.sqrt(2*p*(1-p))\n",
" sd2=mt.sqrt(p*(1-p)+(p+d)*(1-(p+d)))\n",
" sds=[sd1,sd2]\n",
" return sds\n",
"\n",
"# 求Sample Size\n",
"def get_sampSize(sds,alpha,beta,d):\n",
" n=pow((get_z_score(1-alpha/2)*sds[0]+get_z_score(1-beta)*sds[1]),2)/pow(d,2)\n",
" return n"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "uFh5tlyTFyM0"
},
"source": [
"GC[\"d\"]=0.01\n",
"R[\"d\"]=0.01\n",
"NC[\"d\"]=0.0075"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "fiOPWnzNFyM0",
"outputId": "f1b76240-2fe5-44bd-ddcf-225afa73b54a"
},
"source": [
"# Let's get an integer value for simplicity\n",
"GC[\"SampSize\"]=round(get_sampSize(get_sds(GC[\"p\"],GC[\"d\"]),0.05,0.2,GC[\"d\"]))\n",
"GC[\"SampSize\"]"
],
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"25835.0"
]
},
"metadata": {
"tags": []
},
"execution_count": 11
}
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "X7vCHRGDFyM0",
"outputId": "647b76b3-d56e-4dba-e45b-005e919fd854"
},
"source": [
"GC[\"SampSize\"]=round(GC[\"SampSize\"]/0.08*2)\n",
"GC[\"SampSize\"]"
],
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"645875.0"
]
},
"metadata": {
"tags": []
},
"execution_count": 12
}
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"collapsed": true,
"id": "z3jx0jgiFyM0",
"outputId": "1464feca-9b43-43c6-cb6d-aae739d2eb5a"
},
"source": [
"# Getting a nice integer value\n",
"R[\"SampSize\"]=round(get_sampSize(get_sds(R[\"p\"],R[\"d\"]),0.05,0.2,R[\"d\"]))\n",
"R[\"SampSize\"]"
],
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"39087.0"
]
},
"metadata": {
"tags": []
},
"execution_count": 13
}
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"collapsed": true,
"id": "Zc1NFMp3FyM0",
"outputId": "ce917636-2197-45af-a7a5-d75eccc51398"
},
"source": [
"R[\"SampSize\"]=R[\"SampSize\"]/0.08/0.20625*2\n",
"R[\"SampSize\"]"
],
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"4737818.181818182"
]
},
"metadata": {
"tags": []
},
"execution_count": 14
}
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"collapsed": true,
"id": "XqkTq_D-FyM0",
"outputId": "2b1a96db-4bd1-49c9-ac49-02f65f95ee0b"
},
"source": [
"# Getting a nice integer value\n",
"NC[\"SampSize\"]=round(get_sampSize(get_sds(NC[\"p\"],NC[\"d\"]),0.05,0.2,NC[\"d\"]))\n",
"NC[\"SampSize\"]"
],
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"27413.0"
]
},
"metadata": {
"tags": []
},
"execution_count": 15
}
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"collapsed": true,
"id": "UGArSr5lFyM0",
"outputId": "db5dcecd-d9ff-4b29-8ef4-41364a2591bf"
},
"source": [
"NC[\"SampSize\"]=NC[\"SampSize\"]/0.08*2\n",
"NC[\"SampSize\"]"
],
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"685325.0"
]
},
"metadata": {
"tags": []
},
"execution_count": 16
}
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 198
},
"id": "Oj_s62oaFyM1",
"outputId": "b9b4f542-7483-4d29-b598-c27e595f2e2e"
},
"source": [
"# 載入數據\n",
"control=pd.read_csv(\"./sample_data/control_data.csv\")\n",
"experiment=pd.read_csv(\"./sample_data/experiment_data.csv\")\n",
"control.head()"
],
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>Date</th>\n",
" <th>Pageviews</th>\n",
" <th>Clicks</th>\n",
" <th>Enrollments</th>\n",
" <th>Payments</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>Sat, Oct 11</td>\n",
" <td>7723</td>\n",
" <td>687</td>\n",
" <td>134.0</td>\n",
" <td>70.0</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>Sun, Oct 12</td>\n",
" <td>9102</td>\n",
" <td>779</td>\n",
" <td>147.0</td>\n",
" <td>70.0</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>Mon, Oct 13</td>\n",
" <td>10511</td>\n",
" <td>909</td>\n",
" <td>167.0</td>\n",
" <td>95.0</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>Tue, Oct 14</td>\n",
" <td>9871</td>\n",
" <td>836</td>\n",
" <td>156.0</td>\n",
" <td>105.0</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>Wed, Oct 15</td>\n",
" <td>10014</td>\n",
" <td>837</td>\n",
" <td>163.0</td>\n",
" <td>64.0</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" Date Pageviews Clicks Enrollments Payments\n",
"0 Sat, Oct 11 7723 687 134.0 70.0\n",
"1 Sun, Oct 12 9102 779 147.0 70.0\n",
"2 Mon, Oct 13 10511 909 167.0 95.0\n",
"3 Tue, Oct 14 9871 836 156.0 105.0\n",
"4 Wed, Oct 15 10014 837 163.0 64.0"
]
},
"metadata": {
"tags": []
},
"execution_count": 21
}
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"collapsed": true,
"id": "c2AH9yHaFyM1",
"outputId": "5828ee0c-f831-4862-e918-14b3dcf12f28"
},
"source": [
"pageviews_cont=control['Pageviews'].sum()\n",
"pageviews_exp=experiment['Pageviews'].sum()\n",
"pageviews_total=pageviews_cont+pageviews_exp\n",
"print (\"number of pageviews in control:\", pageviews_cont)\n",
"print (\"number of Pageviewsin experiment:\" ,pageviews_exp)"
],
"execution_count": null,
"outputs": [
{
"output_type": "stream",
"text": [
"number of pageviews in control: 345543\n",
"number of Pageviewsin experiment: 344660\n"
],
"name": "stdout"
}
]
},
{
"cell_type": "code",
"metadata": {
"id": "kxKPjYuWFyM1"
},
"source": [
"# Count the total clicks from complete records only\n",
"clicks_cont=control[\"Clicks\"].loc[control[\"Enrollments\"].notnull()].sum()\n",
"clicks_exp=experiment[\"Clicks\"].loc[experiment[\"Enrollments\"].notnull()].sum()"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "I4gtEo6LFyM1",
"outputId": "cb83ee22-d69a-4feb-c6bb-52592a08b077"
},
"source": [
"#Gross Conversion - number of enrollments divided by number of clicks\n",
"enrollments_cont=control[\"Enrollments\"].sum()\n",
"enrollments_exp=experiment[\"Enrollments\"].sum()\n",
"\n",
"GC_cont=enrollments_cont/clicks_cont\n",
"GC_exp=enrollments_exp/clicks_exp\n",
"GC_pooled=(enrollments_cont+enrollments_exp)/(clicks_cont+clicks_exp)\n",
"GC_sd_pooled=mt.sqrt(GC_pooled*(1-GC_pooled)*(1/clicks_cont+1/clicks_exp))\n",
"GC_ME=round(get_z_score(1-alpha/2)*GC_sd_pooled,4)\n",
"GC_diff=round(GC_exp-GC_cont,4)\n",
"print(\"The change due to the experiment is\",GC_diff*100,\"%\")\n",
"print(\"Confidence Interval: [\",GC_diff-GC_ME,\",\",GC_diff+GC_ME,\"]\")\n",
"print (\"The change is statistically significant if the CI doesn't include 0. In that case, it is practically significant if\",-GC[\"d_min\"],\"is not in the CI as well.\")"
],
"execution_count": null,
"outputs": [
{
"output_type": "stream",
"text": [
"The change due to the experiment is -2.06 %\n",
"Confidence Interval: [ -0.0292 , -0.012 ]\n",
"The change is statistically significant if the CI doesn't include 0. In that case, it is practically significant if -0.01 is not in the CI as well.\n"
],
"name": "stdout"
}
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"collapsed": true,
"id": "MQIg2XBsFyM1",
"outputId": "74843a74-15e2-4f60-c0a7-469d7a11acc2"
},
"source": [
"#Net Conversion - number of payments divided by number of clicks\n",
"payments_cont=control[\"Payments\"].sum()\n",
"payments_exp=experiment[\"Payments\"].sum()\n",
"\n",
"NC_cont=payments_cont/clicks_cont\n",
"NC_exp=payments_exp/clicks_exp\n",
"NC_pooled=(payments_cont+payments_exp)/(clicks_cont+clicks_exp)\n",
"NC_sd_pooled=mt.sqrt(NC_pooled*(1-NC_pooled)*(1/clicks_cont+1/clicks_exp))\n",
"NC_ME=round(get_z_score(1-alpha/2)*NC_sd_pooled,4)\n",
"NC_diff=round(NC_exp-NC_cont,4)\n",
"print(\"The change due to the experiment is\",NC_diff*100,\"%\")\n",
"print(\"Confidence Interval: [\",NC_diff-NC_ME,\",\",NC_diff+NC_ME,\"]\")\n",
"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.\")"
],
"execution_count": null,
"outputs": [
{
"output_type": "stream",
"text": [
"The change due to the experiment is -0.49 %\n",
"Confidence Interval: [ -0.0116 , 0.0018000000000000004 ]\n",
"The change is statistically significant if the CI doesn't include 0. In that case, it is practically significant if 0.0075 is not in the CI as well.\n"
],
"name": "stdout"
}
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "K2hnLtKrFS75"
},
"source": [
"# **作業**\n",
"# 經由範例程式碼，熟悉A/B Test的步驟\n",
"\n",
"請同學逐步跟隨程式了解A/B Test步驟"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "-lO_8AYwuEDY"
},
"source": [
"# **作業 嘗試以函數算出樣本數**"
]
},
{
"cell_type": "code",
"metadata": {
"id": "oRSbrNaRuRJA"
},
"source": [
"#作業 Sample Size\n",
"import statsmodels.stats.api as sms\n",
"from math import ceil\n",
"\n",
"effect_size = sms.proportion_effectsize(GC[\"p\"]-1.0*GC[\"d_min\"], GC[\"p\"]+0.0*GC[\"d_min\"])\n",
"required_n = sms.NormalIndPower().solve_power(\n",
" effect_size, \n",
" power=0.8, \n",
" alpha=0.05, \n",
" ratio=1\n",
" ) \n",
"required_n = ceil(required_n) \n",
"print (effect_size,required_n) "
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "MKB09_mjFwjN"
},
"source": [
"# **作業** 自行開發雙樣本比例的信賴區間函數\n"
]
},
{
"cell_type": "code",
"metadata": {
"id": "yT5goD1jHKpl"
},
"source": [
"#作業解答\n",
"import scipy.stats as stats\n",
"def two_proprotions_confint(success_a, size_a, success_b, size_b, significance = 0.05):\n",
" \"\"\"\n",
" A/B test for two proportions;\n",
" given a success a trial size of group A and B compute\n",
" its confidence interval;\n",
" resulting confidence interval matches R's prop.test function\n",
"\n",
" Parameters\n",
" ----------\n",
" success_a, success_b : int\n",
" Number of successes in each group\n",
"\n",
" size_a, size_b : int\n",
" Size, or number of observations in each group\n",
"\n",
" significance : float, default 0.05\n",
" Often denoted as alpha. Governs the chance of a false positive.\n",
" A significance level of 0.05 means that there is a 5% chance of\n",
" a false positive. In other words, our confidence level is\n",
" 1 - 0.05 = 0.95\n",
"\n",
" Returns\n",
" -------\n",
" prop_diff : float\n",
" Difference between the two proportion\n",
"\n",
" confint : 1d ndarray\n",
" Confidence interval of the two proportion test\n",
" \"\"\"\n",
" prop_a = success_a / size_a\n",
" prop_b = success_b / size_b\n",
" var = prop_a * (1 - prop_a) / size_a + prop_b * (1 - prop_b) / size_b\n",
" se = np.sqrt(var)\n",
"\n",
" # z critical value\n",
" confidence = 1 - significance\n",
" z = stats.norm(loc = 0, scale = 1).ppf(confidence + significance / 2)\n",
"\n",
" # standard formula for the confidence interval\n",
" # point-estimtate +- z * standard-error\n",
" prop_diff = prop_b - prop_a\n",
" confint = prop_diff + np.array([-1, 1]) * z * se\n",
" return prop_diff, confint\n",
"two_proprotions_confint(enrollments_cont, clicks_cont, enrollments_exp, clicks_exp, significance = 0.05)"
],
"execution_count": null,
"outputs": []
}
]
}