Mancer1 commented on code in PR #2560:
URL: https://github.com/apache/systemds/pull/2560#discussion_r3770590536
##########
src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupFactory.java:
##########
@@ -1078,6 +1083,46 @@ private static AColGroup
compressLinearFunctional(IColIndex colIndexes, MatrixBl
return ColGroupLinearFunctional.create(colIndexes,
coefficients, numRows);
}
+ /**
+ * This method is the entry point to compress a matrix with piecewise
linear compression The first method uses a
+ * segmented least squares with dynamic programming to compress the
columns The second method uses a successive
+ * compression method, which compares each values in linear time and
checks if the targetloss exceeded
+ *
+ * @param colIndexes the column indices to compress
+ * @param in the input Matrixblock containing the data
+ * @param cs compression settings to define the target loss,
which should be considered
+ * @return a piecewise linear compressed column group
+ */
+
+ public static AColGroup compressPiecewiseLinearFunctional(IColIndex
colIndexes, MatrixBlock in,
+ CompressionSettings cs) {
+
+ final int numRows = in.getNumRows();
+ final int numCols = colIndexes.size();
+ int[][] breakpointsPerCol = new int[numCols][];
+ double[][] slopesPerCol = new double[numCols][];
+ double[][] interceptsPerCol = new double[numCols][];
+
+ for(int col = 0; col < numCols; col++) {
+ final int colIdx = colIndexes.get(col);
+ double[] column = PiecewiseLinearUtils.getColumn(in,
colIdx);
+ PiecewiseLinearUtils.SegmentedRegression fit =
PiecewiseLinearUtils
+ .compressSuccessivePiecewiseLinear(column, cs);
+ breakpointsPerCol[col] = fit.getBreakpoints();
+ interceptsPerCol[col] = fit.getIntercepts();
+ slopesPerCol[col] = fit.getSlopes();
+
+ }
+ return ColGroupPiecewiseLinearCompressed.create(colIndexes,
breakpointsPerCol, slopesPerCol, interceptsPerCol,
+ numRows);
+
+ }
Review Comment:
Done. Just need Tests for coverage
--
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]
