Changeset 33326

Timestamp:

Feb 21, 2012, 4:37:38 PM (14 years ago)

Author:

watersc1

Message:

Octave claims that my inputs on the test image should solve to a
reasonable answer using LU decomposition. Trying the Gauss-Jordan
solver (which should give equivalent results) gives me a solution
vector that agrees with octave to within a constant. Therefore, since
psLib's implementation of GJ seems to be more robust than the
implementation of LU, and the fact that both solvers should yield the
same answer, and the fact that LU is preferred largely due to speed,
and that's not a big issue here, I'm making the switch from LU to GJ.

I'm not terribly satisfied with this, however, as I would have
expected LU and GJ to give the same answers to the same inputs.

File:

• trunk/psModules/src/detrend/pmPattern.c (modified) (1 diff)

trunk/psModules/src/detrend/pmPattern.c

@@ -803 +803 @@
     
     // Solve the Ax=b equation
-    psMatrixLUSolve(XX,solution);
-
+    //    psMatrixLUSolve(XX,solution);
+    psMatrixGJSolve(XX,solution);
+#if (0)
+    for (int i = 0; i < numCells; i++) { // print vector b
+      psTrace("psModules.detrend.cont",5,"b: %d %f",
+              i,
+              solution->data.F64[i]
+              );
+    }
+#endif
+    
     /* old code to remove the minimum solution value from the set, to give a "minimal set of offsets." Mathematically unnecessary. */
 /*     double min = 99e99; */