Created
October 20, 2016 23:16
-
-
Save srnghn/4b10aa0c623a4b246509ffb2e8ac3bb2 to your computer and use it in GitHub Desktop.
ANOVA Test for Spark 2.0 using PySpark. The function returns 5 values: degrees of freedom between (numerator), degrees of freedom within (denominator), F-value, eta squared and omega squared.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
from pyspark.sql.functions import * | |
# Implementation of ANOVA function: calculates the degrees of freedom, F-value, eta squared and omega squared values. | |
# Expects that 'categoryData' with two columns, the first being the categorical independent variable and the second being the scale dependent variable | |
def getAnovaStats(categoryData) : | |
cat_val = categoryData.toDF("cat","value") | |
cat_val.createOrReplaceTempView("df") | |
newdf = spark.sql("select A.cat, A.value, cast((A.value * A.value) as double) as valueSq, ((A.value - B.avg) * (A.value - B.avg)) as diffSq from df A join (select cat, avg(value) as avg from df group by cat) B where A.cat = B.cat") | |
grouped = newdf.groupBy("cat") | |
sums = grouped.sum("value") | |
counts = grouped.count() | |
numCats = counts.count() | |
sumsq = grouped.sum("valueSq") | |
avgs = grouped.avg("value") | |
totN = counts.selectExpr("sum(count) as total").rdd.map(lambda x: x.total).collect()[0] | |
totSum = sums.selectExpr("sum(`sum(value)`) as totSum").rdd.map(lambda x: x.totSum).collect()[0] | |
totSumSq = sumsq.selectExpr("sum(`sum(valueSq)`) as totSumSq").rdd.map(lambda x: x.totSumSq).collect()[0] | |
totMean = totSum / totN | |
dft = totN - 1 | |
dfb = numCats - 1 | |
dfw = totN - numCats | |
joined = counts.selectExpr("cat as category", "count").join(sums, col("category") == sums.cat, 'inner')\ | |
.drop(sums.cat)\ | |
.join(sumsq, col("category") == sumsq.cat, 'inner')\ | |
.drop(sumsq.cat)\ | |
.join(avgs, col("category") == avgs.cat, 'inner')\ | |
.drop(avgs.cat) | |
finaldf = joined.withColumn("totMean", lit(totMean)) | |
ssb_tmp = finaldf.rdd.map(lambda x: (x[0], ((x[4] - x[5])*(x[4] - x[5]))*x[1])) | |
ssb = ssb_tmp.toDF().selectExpr("sum(_2) as total").rdd.map(lambda x: x.total).collect()[0] | |
ssw_tmp = grouped.sum("diffSq") | |
ssw = ssw_tmp.selectExpr("sum(`sum(diffSq)`) as total").rdd.map(lambda x: x.total).collect()[0] | |
sst = ssb + ssw | |
msb = ssb / dfb | |
msw = ssw / dfw | |
F = msb / msw | |
etaSq = ssb / sst | |
omegaSq = (ssb - ((numCats - 1) * msw))/(sst + msw) | |
return (dfb, dfw, F, etaSq, omegaSq) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Which one is the p value?