[MINOR] Increase Memory Estimate for Frames

This commit increases the default estimate of frame size.
Previously, frames were estimated similarly to Matrices.
The wrong estimate leads to problems on frames of > Integer rows.
To improve it, this commit defaults to 8 character strings on all cells.
In an unread matrix.

Since there is no way of knowing if the input Frame contains longer
strings, it is still a subpar estimate. However,
it is an improvement overestimating everything as a dense double Matrix.

(The change happened because I encountered very incorrect estimates in BEWARE)

Author: Baunsgaard

src/main/java/org/apache/sysds/hops/DataOp.java

@@ -359,8 +359,8 @@ public boolean allowsAllExecTypes()
 	protected double computeOutputMemEstimate( long dim1, long dim2, long nnz )
 	{		
 		double ret = 0;
-		
-		if ( getDataType() == DataType.SCALAR ) 
+		final DataType dt = getDataType();
+		if ( dt == DataType.SCALAR ) 
 		{
 			switch( getValueType() ) 
 			{
@@ -379,6 +379,11 @@ protected double computeOutputMemEstimate( long dim1, long dim2, long nnz )
 					ret = 0;
 			}
 		}
+		else if(dt == DataType.FRAME) {
+			if(_op == OpOpData.PERSISTENTREAD || _op == OpOpData.TRANSIENTREAD) {
+				ret = OptimizerUtils.estimateSizeExactFrame(dim1, dim2);
+			}
+		}
 		else //MATRIX / FRAME
 		{
 			if(   _op == OpOpData.PERSISTENTREAD 

src/main/java/org/apache/sysds/hops/OptimizerUtils.java

@@ -64,6 +64,7 @@
 import org.apache.sysds.runtime.util.IndexRange;
 import org.apache.sysds.runtime.util.UtilFunctions;
 import org.apache.sysds.utils.stats.InfrastructureAnalyzer;
+import org.apache.sysds.utils.MemoryEstimates;
 
 public class OptimizerUtils 
 {
@@ -788,6 +789,15 @@ public static long estimateSizeExactSparsity(long nrows, long ncols, long nnz)
 		double sp = getSparsity(nrows, ncols, nnz);
 		return estimateSizeExactSparsity(nrows, ncols, sp);
 	}
+
+
+	public static long estimateSizeExactFrame(long nRows, long nCols){
+		if(nRows > Integer.MAX_VALUE)
+			return Long.MAX_VALUE;
+		
+		// assuming String arrays and on average 8 characters per value.
+		return (long)MemoryEstimates.stringArrayCost((int)nRows, 8) * nCols;
+	}
 
 	/**
 	 * Estimates the footprint (in bytes) for an in-memory representation of a