database/mariadb-columnstore-engine
1
0
Fork 0
mirror of https://github.com/mariadb-corporation/mariadb-columnstore-engine.git synced 2025-08-01 06:46:55 +03:00
Code Activity

Welford algorithm for STD and VAR

Browse Source 
Naive algorithm for calculating STD and VAR is subject to catastrophic
cancellation. A well-known Welford's algorithms is used instead.
This commit is contained in:
Andrey Piskunov
2022-06-01 19:02:24 +03:00
parent 4e50fca460
commit c5fa27475d
8 changed files with 1003 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
dbcon/joblist/tupleaggregatestep.cpp
View File

@ -1478,7 +1478,7 @@ void TupleAggregateStep::prep1PhaseAggregate(JobInfo& jobInfo, vector<RowGroup>&
functionVec[i]->fAuxColumnIndex = lastCol;
// sum(x)
// mean(x)
oidsAgg.push_back(oidsProj[j]);
keysAgg.push_back(keysProj[j]);
scaleAgg.push_back(0);
@ -1488,7 +1488,7 @@ void TupleAggregateStep::prep1PhaseAggregate(JobInfo& jobInfo, vector<RowGroup>&
widthAgg.push_back(sizeof(long double));
++lastCol;
// sum(x**2)
// sum(x_i - mean)^2
oidsAgg.push_back(oidsProj[j]);
keysAgg.push_back(keysProj[j]);
scaleAgg.push_back(0);
@ -1910,7 +1910,7 @@ void TupleAggregateStep::prep1PhaseDistinctAggregate(JobInfo& jobInfo, vector<Ro
widthAgg.push_back(sizeof(double));
funct->fAuxColumnIndex = ++colAgg;
// sum(x)
// mean(x)
oidsAgg.push_back(oidsProj[colProj]);
keysAgg.push_back(aggKey);
scaleAgg.push_back(0);
@ -1920,7 +1920,7 @@ void TupleAggregateStep::prep1PhaseDistinctAggregate(JobInfo& jobInfo, vector<Ro
widthAgg.push_back(sizeof(long double));
++colAgg;
// sum(x**2)
// sum(x_i - mean)^2
oidsAgg.push_back(oidsProj[colProj]);
keysAgg.push_back(aggKey);
scaleAgg.push_back(0);
@ -2581,7 +2581,7 @@ void TupleAggregateStep::prep1PhaseDistinctAggregate(JobInfo& jobInfo, vector<Ro
functionVec2[i]->fAuxColumnIndex = lastCol;
// sum(x)
// mean(x)
oidsAggDist.push_back(oidsAgg[j]);
keysAggDist.push_back(keysAgg[j]);
scaleAggDist.push_back(0);
@ -2591,7 +2591,7 @@ void TupleAggregateStep::prep1PhaseDistinctAggregate(JobInfo& jobInfo, vector<Ro
widthAggDist.push_back(sizeof(long double));
++lastCol;
// sum(x**2)
// sum(x_i - mean)^2
oidsAggDist.push_back(oidsAgg[j]);
keysAggDist.push_back(keysAgg[j]);
scaleAggDist.push_back(0);
@ -3243,7 +3243,7 @@ void TupleAggregateStep::prep2PhasesAggregate(JobInfo& jobInfo, vector<RowGroup>
widthAggPm.push_back(sizeof(double));
funct->fAuxColumnIndex = ++colAggPm;
// sum(x)
// mean(x)
oidsAggPm.push_back(oidsProj[colProj]);
keysAggPm.push_back(aggKey);
scaleAggPm.push_back(0);
@ -3253,7 +3253,7 @@ void TupleAggregateStep::prep2PhasesAggregate(JobInfo& jobInfo, vector<RowGroup>
widthAggPm.push_back(sizeof(long double));
++colAggPm;
// sum(x**2)
// sum(x_i - mean)^2
oidsAggPm.push_back(oidsProj[colProj]);
keysAggPm.push_back(aggKey);
scaleAggPm.push_back(0);
@ -3701,7 +3701,7 @@ void TupleAggregateStep::prep2PhasesAggregate(JobInfo& jobInfo, vector<RowGroup>
functionVecUm[i]->fAuxColumnIndex = lastCol;
// sum(x)
// mean(x)
oidsAggUm.push_back(oidsAggPm[j]);
keysAggUm.push_back(keysAggPm[j]);
scaleAggUm.push_back(0);
@ -3711,7 +3711,7 @@ void TupleAggregateStep::prep2PhasesAggregate(JobInfo& jobInfo, vector<RowGroup>
widthAggUm.push_back(sizeof(long double));
++lastCol;
// sum(x**2)
// sum(x_i - mean)^2
oidsAggUm.push_back(oidsAggPm[j]);
keysAggUm.push_back(keysAggPm[j]);
scaleAggUm.push_back(0);
@ -4152,7 +4152,7 @@ void TupleAggregateStep::prep2PhasesDistinctAggregate(JobInfo& jobInfo, vector<R
widthAggPm.push_back(sizeof(double));
funct->fAuxColumnIndex = ++colAggPm;
// sum(x)
// mean(x)
oidsAggPm.push_back(oidsProj[colProj]);
keysAggPm.push_back(aggKey);
scaleAggPm.push_back(0);
@ -4162,7 +4162,7 @@ void TupleAggregateStep::prep2PhasesDistinctAggregate(JobInfo& jobInfo, vector<R
widthAggPm.push_back(sizeof(long double));
++colAggPm;
// sum(x**2)
// sum(x_i - mean)^2
oidsAggPm.push_back(oidsProj[colProj]);
keysAggPm.push_back(aggKey);
scaleAggPm.push_back(0);
@ -4808,7 +4808,7 @@ void TupleAggregateStep::prep2PhasesDistinctAggregate(JobInfo& jobInfo, vector<R
functionVecUm[i]->fAuxColumnIndex = lastCol;
// sum(x)
// mean(x)
oidsAggDist.push_back(oidsAggPm[j]);
keysAggDist.push_back(keysAggPm[j]);
scaleAggDist.push_back(0);
@ -4818,7 +4818,7 @@ void TupleAggregateStep::prep2PhasesDistinctAggregate(JobInfo& jobInfo, vector<R
widthAggDist.push_back(sizeof(long double));
++lastCol;
// sum(x**2)
// sum(x_i - mean)^2
oidsAggDist.push_back(oidsAggPm[j]);
keysAggDist.push_back(keysAggPm[j]);
scaleAggDist.push_back(0);