optimization and stability

aws · Dec 15, 2023 · b5b17cb · b5b17cb
1 parent eba5608
commit b5b17cb
Show file tree

Hide file tree

Showing 5 changed files with 20 additions and 16 deletions.
diff --git a/Java/core/src/main/java/com/amazon/randomcutforest/RandomCutForest.java b/Java/core/src/main/java/com/amazon/randomcutforest/RandomCutForest.java
@@ -1087,11 +1087,12 @@ public double[] imputeMissingValues(double[] point, int numberOfMissingValues, i
      * @return a forecasted time series.
      */
     @Deprecated
-    public double[] extrapolateBasic(double[] point, int horizon, int blockSize, boolean cyclic, int shingleIndex) {
+    double[] extrapolateBasic(double[] point, int horizon, int blockSize, boolean cyclic, int shingleIndex) {
         return toDoubleArray(extrapolateBasic(toFloatArray(point), horizon, blockSize, cyclic, shingleIndex));
     }
 
-    public float[] extrapolateBasic(float[] point, int horizon, int blockSize, boolean cyclic, int shingleIndex) {
+    @Deprecated
+    float[] extrapolateBasic(float[] point, int horizon, int blockSize, boolean cyclic, int shingleIndex) {
         return extrapolateWithRanges(point, horizon, blockSize, cyclic, shingleIndex, 1.0).values;
     }
 
@@ -1122,7 +1123,9 @@ public RangeVector extrapolateWithRanges(float[] point, int horizon, int blockSi
     // external management of shingle; can function for both internal and external
     // shingling
     // however blocksize has to be externally managed
-    public RangeVector extrapolateFromShingle(float[] shingle, int horizon, int blockSize, double centrality) {
+
+    @Deprecated
+    RangeVector extrapolateFromShingle(float[] shingle, int horizon, int blockSize, double centrality) {
         return extrapolateWithRanges(shingle, horizon, blockSize, isRotationEnabled(),
                 ((int) nextSequenceIndex()) % shingleSize, centrality);
     }
@@ -1142,11 +1145,11 @@ public RangeVector extrapolateFromShingle(float[] shingle, int horizon, int bloc
      * @return a forecasted time series.
      */
     @Deprecated
-    public double[] extrapolateBasic(double[] point, int horizon, int blockSize, boolean cyclic) {
-        return extrapolateBasic(point, horizon, blockSize, cyclic, 0);
+    double[] extrapolateBasic(double[] point, int horizon, int blockSize, boolean cyclic) {
+        return toDoubleArray(extrapolateBasic(toFloatArray(point), horizon, blockSize, cyclic, 0));
     }
 
-    public float[] extrapolateBasic(float[] point, int horizon, int blockSize, boolean cyclic) {
+    protected float[] extrapolateBasic(float[] point, int horizon, int blockSize, boolean cyclic) {
         return extrapolateBasic(point, horizon, blockSize, cyclic, 0);
     }
 
@@ -1162,8 +1165,8 @@ public float[] extrapolateBasic(float[] point, int horizon, int blockSize, boole
      */
     @Deprecated
     public double[] extrapolateBasic(ShingleBuilder builder, int horizon) {
-        return extrapolateBasic(builder.getShingle(), horizon, builder.getInputPointSize(), builder.isCyclic(),
-                builder.getShingleIndex());
+        return toDoubleArray(extrapolateBasic(toFloatArray(builder.getShingle()), horizon, builder.getInputPointSize(),
+                builder.isCyclic(), builder.getShingleIndex()));
     }
 
     void extrapolateBasicSliding(RangeVector result, int horizon, int blockSize, float[] queryPoint,
@@ -1180,12 +1183,11 @@ void extrapolateBasicSliding(RangeVector result, int horizon, int blockSize, flo
             System.arraycopy(queryPoint, blockSize, queryPoint, 0, dimensions - blockSize);
 
             SampleSummary imputedSummary = getConditionalFieldSummary(queryPoint, missingIndexes, 1, 0, false, false,
-                    centrality, 1);
+                    centrality, dimensions / blockSize);
             for (int y = 0; y < blockSize; y++) {
-                result.values[resultIndex] = queryPoint[dimensions - blockSize + y] = imputedSummary.median[dimensions
-                        - blockSize + y];
-                result.lower[resultIndex] = imputedSummary.lower[dimensions - blockSize + y];
-                result.upper[resultIndex] = imputedSummary.upper[dimensions - blockSize + y];
+                result.values[resultIndex] = queryPoint[dimensions - blockSize + y] = imputedSummary.median[y];
+                result.lower[resultIndex] = imputedSummary.lower[y];
+                result.upper[resultIndex] = imputedSummary.upper[y];
                 resultIndex++;
             }
         }

diff --git a/Java/core/src/test/java/com/amazon/randomcutforest/preprocessor/PreprocessorTest.java b/Java/core/src/test/java/com/amazon/randomcutforest/preprocessor/PreprocessorTest.java
@@ -244,8 +244,8 @@ public void preprocessorPlusForest(int seed, ForestMode mode, TransformMethod me
                     preprocessor.update(dataWithKey.data[i], input, timestamp, new int[1], forest);
                 }
                 if (shingleSize > 1) {
-                    RangeVector rangeVector = forest.extrapolateFromShingle(preprocessor.getLastShingledPoint(), 1,
-                            tempDimensions / shingleSize, 1.0);
+                    RangeVector rangeVector = forest.extrapolateWithRanges(preprocessor.getLastShingledPoint(), 1,
+                            tempDimensions / shingleSize, false, 0, 1.0);
                     TimedRangeVector timedRanges = preprocessor.invertForecastRange(rangeVector, timestamp, null, false,
                             timestamp);
                     // error of lookahead

diff --git a/...ces/src/main/java/com/amazon/randomcutforest/parkservices/ThresholdedRandomCutForest.java b/...ces/src/main/java/com/amazon/randomcutforest/parkservices/ThresholdedRandomCutForest.java
@@ -383,7 +383,7 @@ public TimedRangeVector extrapolate(int horizon, boolean correct, double central
                             preprocessor.getScale(), transformMethod, lastAnomalyDescriptor);
                 }
             }
-            RangeVector answer = forest.extrapolateFromShingle(newPoint, horizon, blockSize, centrality);
+            RangeVector answer = forest.extrapolateWithRanges(newPoint, horizon, blockSize, false, 0, centrality);
             return preprocessor.invertForecastRange(answer, lastAnomalyDescriptor.getInputTimestamp(),
                     lastAnomalyDescriptor.getDeltaShift(), lastAnomalyDescriptor.getExpectedRCFPoint() != null,
                     lastAnomalyDescriptor.getExpectedTimeStamp());

diff --git a/...services/src/main/java/com/amazon/randomcutforest/parkservices/state/RCFCasterMapper.java b/...services/src/main/java/com/amazon/randomcutforest/parkservices/state/RCFCasterMapper.java
@@ -91,6 +91,7 @@ public RCFCaster toModel(RCFCasterState state, long seed) {
         descriptor.setTransformMethod(transformMethod);
         descriptor
                 .setImputationMethod(ImputationMethod.valueOf(state.getPreprocessorStates()[0].getImputationMethod()));
+        descriptor.setShingleSize(preprocessor.getShingleSize());
 
         PredictorCorrectorMapper mapper = new PredictorCorrectorMapper();
         PredictorCorrector predictorCorrector = mapper.toModel(state.getPredictorCorrectorState());

diff --git a/.../java/com/amazon/randomcutforest/parkservices/state/ThresholdedRandomCutForestMapper.java b/.../java/com/amazon/randomcutforest/parkservices/state/ThresholdedRandomCutForestMapper.java
@@ -76,6 +76,7 @@ public ThresholdedRandomCutForest toModel(ThresholdedRandomCutForestState state,
             descriptor = new ComputeDescriptorMapper().toModel(state.getLastDescriptorState());
         }
 
+        descriptor.setShingleSize(preprocessor.getShingleSize());
         descriptor.setForestMode(forestMode);
         descriptor.setTransformMethod(transformMethod);
         descriptor.setScoringStrategy(scoringStrategy);