xiaohui-sun commented on a change in pull request #4960: [TE] Use upper and
lower bounds as predicted for threshold algorithm
URL: https://github.com/apache/incubator-pinot/pull/4960#discussion_r363473207
##########
File path:
thirdeye/thirdeye-pinot/src/main/java/org/apache/pinot/thirdeye/detection/components/ThresholdRuleDetector.java
##########
@@ -101,15 +100,24 @@ public TimeSeries computePredictedTimeSeries(MetricSlice
slice) {
InputData data =
this.dataFetcher.fetchData(new
InputDataSpec().withTimeseriesSlices(Collections.singletonList(slice)));
DataFrame df = data.getTimeseries().get(slice);
- return TimeSeries.fromDataFrame(constructThresholdBoundaries(df));
+ return TimeSeries.fromDataFrame(constructBaselineAndBoundaries(df));
}
- private DataFrame constructThresholdBoundaries(DataFrame df) {
+ /**
+ * Populate the dataframe with upper/lower boundaries and baseline
+ */
+ private DataFrame constructBaselineAndBoundaries(DataFrame df) {
+ // Set default baseline as the actual value
+ df.addSeries(COL_VALUE, df.getDoubles(COL_CURRENT));
if (!Double.isNaN(this.min)) {
df.addSeries(COL_LOWER_BOUND, DoubleSeries.fillValues(df.size(),
this.min));
+ // set baseline value as the lower bound when actual value across below
the mark
+ df.mapInPlace(DoubleSeries.MAX, COL_VALUE, COL_LOWER_BOUND, COL_VALUE);
Review comment:
Why we want to change the baseline value?
If it is threshold algorithm we should not show upper/lower bounds.
Please take a look at this doc:
https://docs.google.com/document/d/12bE6e_fIbk3pp0oPCrOHOPFid4NEMguZ-dF3IVjF3xQ/edit
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
