Hk from kalman gain storage #417
Conversation
…g observations, add option to xml
…g observations, take offset into account, add logging and improve code
…g observations, take offset into account
…value for missing observations in Steady State filtering
…value for missing observations in Steady State filtering, move code to separate inner class
…ovation value for missing observations in Steady State filtering, move code to separate inner class" This reverts commit 896c649.
…value for missing observations in Steady State filtering, move code to separate inner class
…value for missing observations in Steady State filtering, move code to separate inner class
…ssociation/OpenDA into HKFromKalmanGainStorage
…rage # Conflicts: # core/java/src/org/openda/utils/io/KalmanGainStorage.java # core/java/test/org/openda/utils/io/KalmanGainStorageTest.java
…FromKalmanGainStorage # Conflicts: # algorithms/java/src/org/openda/algorithms/kalmanFilter/EnKF.java # core/java/src/org/openda/blackbox/wrapper/BBStochModelInstance.java # core/java/src/org/openda/interfaces/IModelInstance.java
nilsvanvelzen
left a comment
nilsvanvelzen
left a comment
There was a problem hiding this comment.
Can you please squash you PR first because
- we do not want to have all these commits in main
- it makes reviewing very difficult
| for(int i=0;i<m;i++){ | ||
| ArrayList<Integer> missingObservationIndices = new ArrayList<>(); | ||
| ArrayList<Integer> availableObservationIndices = new ArrayList<>(); | ||
| for (int i = 0; i < gainStorageObservationIdsArray.length; i++) { |
There was a problem hiding this comment.
You mentioned that documentation will be part of the repository. I suggest to briefly explain what is going on + a link to the detailed description in the document.
| // find matching column in steady-state gain | ||
| String gainVectorId = obsIds[i]+":"+Math.round(obsTimeOffsets[i]*24.0*3600.0); //conversion to seconds | ||
| Results.putProgression("processing obs "+gainVectorId+"\n"); | ||
| String gainVectorId = obsIds[i] + ":" + Math.round(obsTimeOffsets[i] * 24.0 * 3600.0); //conversion to seconds |
There was a problem hiding this comment.
This was part of the old code as well. But it puzzles me a bit what this stuff with conversion to seconds is here. You could add a line of comment to explain what kind of ID we are going to construct.
| private double[] readGainTime; | ||
| private int steadyStateTimeCounter = 0; | ||
| private double skipAssimilationStandardDeviationFactor = Double.POSITIVE_INFINITY; | ||
| private boolean tryCompensatingForMissingObservationWithHK = false; |
There was a problem hiding this comment.
this is a long name :-) What is HK in this context? H observation interpolation and K the gain? Not clear to me.
There was a problem hiding this comment.
I'll ask Julius and Martin if they can give me a better description of HK for inspiration
There was a problem hiding this comment.
Suggestion: estimateMissingObservation
| </xs:element> | ||
| <xs:element name="tryCompensatingForMissingObservationWithHK" type="xs:boolean" minOccurs="0"> | ||
| <xs:annotation> | ||
| <xs:documentation>When this option is true AND in the kalman gain storage HK has been stored, for locations with missing observations the innovation value </xs:documentation> |
There was a problem hiding this comment.
again that "HK" is too cryptical for me (might find the answer later)
There was a problem hiding this comment.
I'll ask Julius and Martin if they can give me a better description of HK for inspiration
|
I'm having difficulties squashing commits, I'll create a new branch and PR to see if that will resolve the issue. |
Implementation of calculating Kalman Gain for stations with missing observations.
In the document below the developments are explained with mathematical background and references to the code:
SSKF_compensate_for_missing_observations.docx