How to run

[zuowang]

Is this project ready? I follow the instructions in README, but I found the code in *.so file is not executed.
Could you tell me how to run the three algorithms, do I need extra libraries? Thanks very much!

[bherta]

The code in this repository is used by the modified Spark in the SparkGPU repository. The code here is mostly intended for Spark's use, but it can also be used by other programs. Please see the READMEs under each of the three algorithm folders.

[zuowang]

I found the code in CUDA-MLlib repository is not compatible with SparkGPU repository
Take ALS as an example:
The scala code want native code implementation for factorize(), getUser(), getProd()
class gpuALS {

@Native def factorize(rank: Int,
num_user: Int,
num_prod: Int,
num_rating: Int,
row_idx: Array[Int],
col_idx: Array[Int],
rating: Array[Float],
lambda: Double, num_itr: Int)

@Native def getUser(): Array[Float]
@Native def getProd(): Array[Float]

}

while the native code in CUDA-MLlib implement doALSWithCSR():

JNIEXPORT jobjectArray JNICALL Java_org_apache_spark_ml_recommendation_CuMFJNIInterface_doALSWithCSR (JNIEnv *, jobject, jint, jint, jint, jint, jdouble, jobjectArray, jintArray, jintArray, jfloatArray);

Could you please update the code in CUDA-MLlib? Thank you so much!

[wei-tan]

Please take a look at https://github.com/IBMSparkGPU/SparkGPU/blob/MLlib/mllib/src/main/scala/org/apache/spark/ml/recommendation/ALS.scala

val jni = new CuMFJNIInterface
dstFactors = jni.doALSWithCSR(dstIds.length, accumulatedSortedSrcFactorLength.last, rank,
ratings.length, regParam, sortedSrcFactors, srcPtrs, csrCol, ratings)

This scala code is coherent with the native implementation.