Merge implementation of MergingDigest

[rnataf]

Hi, I have a question about the implementation of the merge function of MergingDigest: private void merge(double[] incomingMean, double[] incomingWeight, int incomingCount, List<List<Double>> incomingData, int[] incomingOrder, double unmergedWeight, boolean runBackwards, double compression). In the loop for (int i = 1; i < incomingCount; i++) , what prevents the variable lastUsedCell not to increase too much, leading to an illegal access to the array weight[lastUsedCell] ?

I'm asking because I did a small modification to the code (multiplying weights by a constant before continuing adding data, to the object) and it caused lastUsedCell to have a bigger value than weight.length.

[tdunning]

Normally what prevents this is the scale invariant. Each of the current scale functions K0 through K3 has an associated proof that simple merging then centroids of two digests that have the invariant will result in a digest that satisfies the invariant. Further, there is a proof that fully merging any digest that meets the invariant will have a well bounded size. If you multiply the weights, then you very likely wind up with a digest that doesn't meet the invariant and that means that all bets are off.

…

On Wed, Sep 14, 2022 at 5:18 AM rnataf ***@***.***> wrote: Hi, I have a question about the implementation of the merge function of MergingDigest. private void merge(double[] incomingMean, double[] incomingWeight, int incomingCount, List<List<Double>> incomingData, int[] incomingOrder, double unmergedWeight, boolean runBackwards, double compression). In the loop for (int i = 1; i < incomingCount; i++) , what prevents the variable lastUsedCell not to increase too much, leading to an illegal access to the array weight[lastUsedCell]. I'm asking because I did a small modification to the code (multiplying weights by a constant before continuing adding data, to the object) and it caused lastUsedCell to have a bigger value than weight.length. — Reply to this email directly, view it on GitHub <#200>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAB5E6V2JK2MNN4ISFS2ADLV6G7CFANCNFSM6AAAAAAQMLO4SM> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[rnataf]

Okay, thank you for your answer.