[SPARK-29686][ML] LinearSVC should persist instances if needed

### What changes were proposed in this pull request?
persist the input if needed

### Why are the changes needed?
training with non-cached dataset will hurt performance

### Does this PR introduce any user-facing change?
No

### How was this patch tested?
existing tests

Closes apache#26344 from zhengruifeng/linear_svc_cache.

Authored-by: zhengruifeng <[email protected]>
Signed-off-by: zhengruifeng <[email protected]>

Author: zhengruifeng

mllib/src/main/scala/org/apache/spark/ml/classification/LinearSVC.scala

@@ -37,6 +37,7 @@ import org.apache.spark.ml.util.Instrumentation.instrumented
 import org.apache.spark.mllib.linalg.VectorImplicits._
 import org.apache.spark.mllib.stat.MultivariateOnlineSummarizer
 import org.apache.spark.sql.{Dataset, Row}
+import org.apache.spark.storage.StorageLevel
 
 /** Params for linear SVM Classifier. */
 private[classification] trait LinearSVCParams extends ClassifierParams with HasRegParam
@@ -159,7 +160,10 @@ class LinearSVC @Since("2.2.0") (
   override def copy(extra: ParamMap): LinearSVC = defaultCopy(extra)
 
   override protected def train(dataset: Dataset[_]): LinearSVCModel = instrumented { instr =>
+    val handlePersistence = dataset.storageLevel == StorageLevel.NONE
+
     val instances = extractInstances(dataset)
+    if (handlePersistence) instances.persist(StorageLevel.MEMORY_AND_DISK)
 
     instr.logPipelineStage(this)
     instr.logDataset(dataset)
@@ -268,6 +272,8 @@ class LinearSVC @Since("2.2.0") (
       (Vectors.dense(coefficientArray), intercept, scaledObjectiveHistory.result())
     }
 
+    if (handlePersistence) instances.unpersist()
+
     copyValues(new LinearSVCModel(uid, coefficientVector, interceptVector))
   }
 }