Question

我进行了A / B测试，我的数据如下：

  control_conversion  test_conversion  
day1  0.10                0.10  
day3  0.14                0.20
day5  0.20                0.32
day7  0.40                0.80

控制和测试组的流量为1000 因此转换率应为：

control = [0.1, 0.14, 0.20, 0.40]
test = [0.1,0.2,0.32,0.8]

我想使用python来计算第1天，第3天，第5天，第7天用于控制和测试。

所以我需要列出两个清单：

abstract class MainDatabase: RoomDatabase() {
    companion object {
        val instance: MainDatabase by lazy {
            if (_instance == null) throw IllegalStateException("someone should have called init fun")
            _instance!!
        }

        private var _instance: MainDatabase? = null
        fun init(mainApplication: MainApplication) {
            _instance = init_(mainApplication)
            //force db opening and if it fails, we try to destroy and recreate the db
            try {
                _instance!!.openHelper.writableDatabase
            } catch (e: Exception) {
                Log.e("Database", "there was an error during DB opening => trying to destroy and recreate", e)
                _instance!!.openHelper.close()
                val dbPath = mainApplication.getDatabasePath(DB_NAME)
                if (SQLiteDatabase.deleteDatabase(dbPath)) {
                    _instance = init_(mainApplication)
                    _instance!!.openHelper.writableDatabase
                }
            }
        }

        private fun init_(mainApplication: MainApplication): MainDatabase {
            return Room.databaseBuilder(mainApplication, MainDatabase::class.java, DB_NAME)
                    .addMigrations(MIGRATION_1, MIGRATION_2, MIGRATION_3, MIGRATION_4, MIGRATION_5, MIGRATION_6, MIGRATION_7, MIGRATION_8, MIGRATION_9, MIGRATION_10)
                    .build()
        }
    }

如何为两个列表计算四个p值？

就像我想看到的是p值的列表

pvalue = [0.1，0.2，0,1,0.2,0.1]

Answer 1

快速且肮脏，假设控制和测试包含相同数量的项目：

control = [0.1, 0.14, 0.20, 0.40]
test = [0.1,0.2,0.32,0.8]

for idx in range(len(control)):
    val_co = control[idx]
    val_te = test[idx]
    # do whatever you want to do with val_co and val_te

Answer 2

您可以尝试在SciPy中使用binomial test function

from scipy.stats import binom_test
n = 1000

control = [100, 140,200,400]
test = [101,200,320,800]
pvals = []
for idx in range(len(control)):
    pvals.append(binom_test(test[idx],n=n, p=control[idx]/n))

print(pvals)

[0.9160130517865064, 1.8593423831091924e-07, 4.004795877115897e-19, 1.644604962019165e-147]

（我刚刚对此做了basic 101 blog post）

如何通过使用python在ab测试中计算两个列表的统计显着性

2 个答案: