xiaoa6435 opened a new pull request, #42431:
URL: https://github.com/apache/spark/pull/42431
### What changes were proposed in this pull request?
don't flush when ties size exceed 10_000_000, the old behavior results wrong
average rank
### Why are the changes needed?
when data has huge amount of Ties( > 10_000_000), old implementation give
incorrect and inconsistent result. for example
#### scala
```scala
import org.apache.spark.ml.linalg.{Matrix, Vectors, Vector}
import org.apache.spark.ml.stat.Correlation
import org.apache.spark.sql.Row
val N = 10000002
val x = sc.range(0, N).map(i => if (i < N - 1) 1.0 else 2.0)
val y = sc.range(0, N).map(i => if (i < N - 1) 2.0 else 1.0)
//val s1 = Statistics.corr(x, y, "spearman")
val df = x.zip(y)
.map{case (x, y) => Vectors.dense(x, y)}
.map(Tuple1.apply)
.repartition(1)
.toDF("features")
val Row(coeff1: Matrix) = Correlation.corr(df, "features", "spearman").head
val r = coeff1(0, 1)
println(s"pearson correlation in spark: $r")
// pearson correlation in spark: -9.999990476024495E-8
```
current implementation result is -9.999990476024495E-8(unstable), and
correct result is -1.0(in R/python and manual calculation)
#### r
```r
N = 10000002
x = ifelse(0:(N - 1) < N - 1, 1.0, 2.0)
y = ifelse(0:(N - 1) < N - 1, 2.0, 1.0)
r = cor(x, y, method = 'spearman')
sprintf("pearson correlation in r: %0.6f", r)
# pearson correlation in r: -1.000000
```
#### python
```python
import numpy as np
from scipy import stats
N = 10000002
x = np.array(list(1.0 if i < N - 1 else 2.0 for i in range(N)))
y = np.array(list(2.0 if i < N - 1 else 1.0 for i in range(N)))
r = stats.spearmanr(x, y).correlation
print(f"pearson correlation in python scipy.stats: {r}")
# pearson correlation in python scipy.stats: -1.0
```
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
add new test
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
