zzh
/
gatk-3.8
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Quick optimization to the PairHMM 

Problem
--------
the logless HMM scale factor (to avoid double under-flows) was 10^300. Although this serves the purpose this value results in a complex mantissa that further complicates cpu calculations.

Solution
---------
initialize with 2^1020 (2^1023 is the max value), and adjust the scale factor accordingly.
This commit is contained in:
Mauricio Carneiro 2013-04-14 23:25:33 -04:00
parent 55547b68bb
commit 9bfa5eb70f
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
protected/java/src/org/broadinstitute/sting/utils/pairhmm/LoglessPairHMM.java
View File

@ -56,8 +56,8 @@ import org.broadinstitute.sting.utils.QualityUtils;
* Date: 10/16/12
*/
public final class LoglessPairHMM extends PairHMM {
protected static final double SCALE_FACTOR_LOG10 = 300.0;
protected static final double INITIAL_CONDITION = Math.pow(10, SCALE_FACTOR_LOG10);
protected static final double INITIAL_CONDITION = Math.pow(2, 1020);
protected static final double INITIAL_CONDITION_LOG10 = Math.log10(INITIAL_CONDITION);
private static final int matchToMatch = 0;
private static final int indelToMatch = 1;
@ -118,7 +118,7 @@ public final class LoglessPairHMM extends PairHMM {
for (int j = 1; j < paddedHaplotypeLength; j++) {
finalSumProbabilities += matchMatrix[endI][j] + insertionMatrix[endI][j];
}
return Math.log10(finalSumProbabilities) - SCALE_FACTOR_LOG10;
return Math.log10(finalSumProbabilities) - INITIAL_CONDITION_LOG10;
}
/**