NOTICE: This version of the NSF Unidata web site (archive.unidata.ucar.edu) is no longer being updated.
Current content can be found at unidata.ucar.edu.
To learn about what's going on, see About the Archive Site.
NOTICE: This version of the NSF Unidata web site (archive.unidata.ucar.edu) is no longer being updated.
Current content can be found at unidata.ucar.edu.
To learn about what's going on, see About the Archive Site.
Hi Ian, As you say, your data causes the Watson algorithm to fail, so its weird (it includes lots of co-linear points). The Clarkson algorithm doesn't crash, but produces a topology where two triangles get into a tight loop in the search in the interpolation algorithm. However, I have found a "hack" to avoid the tight loop, that seems to make interpolation work with your data. There is no gaurantee that this will work with similar problem data sets. Also, any Delaunay algorithm will be very slow when applied to data sets much larger than yours. The fix is in the attached visad/Irregular2DSet.java. By the way, you last message never made it to visad-list because of the 40 KB limit, but hopefully this reply will. Good luck, Bill ---------------------------------------------------------- Bill Hibbard, SSEC, 1225 W. Dayton St., Madison, WI 53706 hibbard@xxxxxxxxxxxxxxxxx 608-263-4427 fax: 608-263-6738 http://www.ssec.wisc.edu/~billh/vis.html On Tue, 29 Apr 2003, Ian Graham wrote: > > Hi Bill, thanks for your reply. I've attached code that demonstrates the > problems. > > > We will need a test case for debugging this. But first a > > couple questions: > > > > 1. Are you generating your own topology passed to > > DelaunayCustom, or are you letting the system compute > > its own? > > As you'll see in the code, I'm using the default triangulation with 2904 > points, > so it's initially trying the DelaunayWatson, which fails, and then it uses > DelaunayClarkson. > > I _would_ like to understand where the faster algorithms fail, however, > because this is a very small dataset in my world, and I don't need > precision. I already make sure I don't have identical x,y coordinates, but > that doesn't seem to be enough, and I thought only the Clarkson algorithm > rounds to integers. > > > 2. When you offset X and y values, do you do that before > > you or the system computes the topology, or do you compute > > the topology and then offset the x and y values? > > I was offsetting the values before computing the topology, but as it turns > out I'm having trouble repeating the cases where the problem goes away. But > you should find it easy to see the two cases where either part or all of the > resampled data is missing. > > > In any case, we will need a program, any necessary data > > files, and instructions for compiling and running the > > program. > > I've attached the program and data files. Extract them into and empty > folder, and assuming the current directory and visad.jar are in your > classpath you can compile and run as below: > > javac ResamplingProblem.java > java ResamplingProblem > > I hope the problem is easy to find and fix! > > Thanks, > Ian >
//
// Irregular2DSet.java
//
/*
VisAD system for interactive analysis and visualization of numerical
data. Copyright (C) 1996 - 2002 Bill Hibbard, Curtis Rueden, Tom
Rink, Dave Glowacki, Steve Emmerson, Tom Whittaker, Don Murray, and
Tommy Jasmin.
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Library General Public
License as published by the Free Software Foundation; either
version 2 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Library General Public License for more details.
You should have received a copy of the GNU Library General Public
License along with this library; if not, write to the Free
Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
MA 02111-1307, USA
*/
package visad;
/**
* <P>{@link IrregularSet} for a finite number of samples of Rē.</P>
*
* <P>NOTE: There is no {@link Irregular2DSet} with a manifold dimension equal
* to one. Use {@link Gridded2DSet} with a manifold dimension equal to one
* instead.</P>
*
* <p>When you call an {@link Irregular2DSet} constructor without a {@link
* Delaunay} argument, the constructor uses the {@link Delaunay#factory()}
* method to implictly compute a Delaunay triangulation. 3000 points is the
* current break-point from Watson's algorithm to Clarkson's algorithm. So,
* currently, at 3001 points you start using Clarkson's algorithm, which rounds
* coordinates to integers. If your values are small enough that integer
* rounding will merge some of them to the same value (and hence create
* colinear or colocated points), there will be trouble. One approach is
* to scale your coordinates up so integer rounding does not merge values.
* Another is to ensure that you use Watson's algorithm by using <code>new
* DelaunayWatson(samples)</code> as the {@link Delaunay} argument of the {@link
* Irregular2DSet} constructor.</p>
*/
public class Irregular2DSet extends IrregularSet {
private float LowX, HiX, LowY, HiY;
/** a 2-D irregular set with null errors, CoordinateSystem
and Units are defaults from type; topology is computed
by the constructor */
public Irregular2DSet(MathType type, float[][] samples)
throws VisADException {
this(type, samples, null, null, null, null, true);
}
/** a 2-D irregular set; samples array is organized
float[2][number_of_samples]; no geometric constraint on
samples; if delan is non-null it defines the topology of
samples (which must have manifold dimension 2), else the
constructor computes a topology with manifold dimension 2;
note that Gridded2DSet can be used for an irregular set
with domain dimension 2 and manifold dimension 1;
coordinate_system and units must be compatible with
defaults for type, or may be null; errors may be null */
public Irregular2DSet(MathType type, float[][] samples,
CoordinateSystem coord_sys, Unit[] units,
ErrorEstimate[] errors, Delaunay delan)
throws VisADException {
this(type, samples, coord_sys, units, errors, delan, true);
}
public Irregular2DSet(MathType type, float[][] samples,
CoordinateSystem coord_sys, Unit[] units,
ErrorEstimate[] errors, Delaunay delan,
boolean copy) throws VisADException {
super(type, samples, samples.length, coord_sys,
units, errors, delan, copy);
if (samples.length != 2) {
throw new SetException("Irregular2DSet: ManifoldDimension " +
"must be 2 for this constructor");
}
LowX = Low[0];
HiX = Hi[0];
LowY = Low[1];
HiY = Hi[1];
oldToNew = null;
newToOld = null;
}
/* shortcut constructor for constructing Irregular2DSet
using Delaunay from existing Irregular2DSet */
/* CTR: 1-12-98
public Irregular2DSet(MathType type, float[][] samples,
Irregular2DSet delaunay_set) throws VisADException {
this(type, samples, delaunay_set, null, null, null, true);
}
*/
/* complete constructor for constructing Irregular2DSet
using Delaunay from existing Irregular2DSet */
/* CTR: 1-12-98
public Irregular2DSet(MathType type, float[][] samples,
Irregular2DSet delaunay_set,
CoordinateSystem coord_sys, Unit[] units,
ErrorEstimate[] errors) throws VisADException {
this(type, samples, delaunay_set, coord_sys, units, errors, true);
}
public Irregular2DSet(MathType type, float[][] samples,
Irregular2DSet delaunay_set,
CoordinateSystem coord_sys, Unit[] units,
ErrorEstimate[] errors, boolean copy)
throws VisADException {
super(type, samples, delaunay_set.getManifoldDimension(),
coord_sys, units, errors, copy);
int dim = delaunay_set.getManifoldDimension();
if (dim != 2) {
throw new SetException("Irregular2DSet: delaunay_set ManifoldDimension " +
"must be 2");
}
if (Length != delaunay_set.Length) {
throw new SetException("Irregular2DSet: delaunay_set length not match");
}
Delan = delaunay_set.Delan;
LowX = Low[0];
HiX = Hi[0];
LowY = Low[1];
HiY = Hi[1];
oldToNew = null;
newToOld = null;
}
*/
/** shortcut constructor for constructing Irregular2DSet
using sort from existing Irregular1DSet */
public Irregular2DSet(MathType type, float[][] samples,
int[] new2old, int[] old2new) throws VisADException {
this(type, samples, new2old, old2new, null, null, null, true);
}
/** complete constructor for constructing Irregular2DSet
using sort from existing Irregular1DSet */
public Irregular2DSet(MathType type, float[][] samples,
int[] new2old, int[] old2new,
CoordinateSystem coord_sys, Unit[] units,
ErrorEstimate[] errors) throws VisADException {
this(type, samples, new2old, old2new, coord_sys, units, errors, true);
}
public Irregular2DSet(MathType type, float[][] samples,
int[] new2old, int[] old2new,
CoordinateSystem coord_sys, Unit[] units,
ErrorEstimate[] errors, boolean copy)
throws VisADException {
super(type, samples, 1, coord_sys, units, errors, null, copy);
if (Length != new2old.length || Length != old2new.length) {
throw new SetException("Irregular2DSet: sort lengths do not match");
}
newToOld = new int[Length];
oldToNew = new int[Length];
System.arraycopy(new2old, 0, newToOld, 0, Length);
System.arraycopy(old2new, 0, oldToNew, 0, Length);
LowX = Low[0];
HiX = Hi[0];
LowY = Low[1];
HiY = Hi[1];
Delan = null;
}
public Set makeSpatial(SetType type, float[][] samples) throws VisADException
{
if (samples.length == 3) {
if (ManifoldDimension == 1) {
return new Irregular3DSet(type, samples, newToOld, oldToNew,
null, null, null, false);
}
else {
/* WLH 15 Dec 98
if (Delan.Tri == null || Delan.Tri.length == 0) return null;
*/
if (Delan == null || Delan.Tri == null || Delan.Tri.length == 0) return
null;
return new Irregular3DSet(type, samples, null, null, null,
Delan, false);
}
}
else if (samples.length == 2) {
if (ManifoldDimension == 1) {
return new Irregular2DSet(type, samples, newToOld, oldToNew,
null, null, null, false);
}
else {
/* WLH 15 Dec 98
if (Delan.Tri == null || Delan.Tri.length == 0) return null;
*/
if (Delan == null || Delan.Tri == null || Delan.Tri.length == 0) return
null;
return new Irregular2DSet(type, samples, null, null, null,
Delan, false);
}
}
else {
throw new SetException("Irregular2DSet.makeSpatial: bad samples length");
}
}
/** convert an array of 1-D indices to an array of values in
R^DomainDimension */
public float[][] indexToValue(int[] index) throws VisADException {
float[][] value = new float[2][index.length];
for (int i=0; i<index.length; i++) {
if ( (index[i] >= 0) && (index[i] < Length) ) {
value[0][i] = Samples[0][index[i]];
value[1][i] = Samples[1][index[i]];
}
else {
value[0][i] = value[1][i] = Float.NaN;
}
}
return value;
}
/** valueToTri returns an array of containing triangles given
an array of points in R^DomainDimension */
public int[] valueToTri(float[][] value) throws VisADException {
if (ManifoldDimension != 2) {
throw new SetException("Irregular2DSet.valueToTri: " +
"ManifoldDimension must be 2, not " +
ManifoldDimension);
}
int length = value[0].length;
if (length != value[1].length) {
throw new SetException("Irregular2DSet.valueToTri: lengths " +
"don't match");
}
int[] tri = new int[length];
int curtri = 0;
for (int i=0; i<length; i++) {
// Return -1 if iteration loop fails
tri[i] = -1;
boolean foundit = false;
if (curtri < 0) curtri = 0;
for (int itnum=0; (itnum<Delan.Tri.length) && !foundit; itnum++) {
// define data
int t0 = Delan.Tri[curtri][0];
int t1 = Delan.Tri[curtri][1];
int t2 = Delan.Tri[curtri][2];
float Ax = Samples[0][t0];
float Ay = Samples[1][t0];
float Bx = Samples[0][t1];
float By = Samples[1][t1];
float Cx = Samples[0][t2];
float Cy = Samples[1][t2];
float Px = value[0][i];
float Py = value[1][i];
// tests whether point is contained in current triangle
float tval0 = (Bx-Ax)*(Py-Ay) - (By-Ay)*(Px-Ax);
float tval1 = (Cx-Bx)*(Py-By) - (Cy-By)*(Px-Bx);
float tval2 = (Ax-Cx)*(Py-Cy) - (Ay-Cy)*(Px-Cx);
boolean test0 = (tval0 == 0) || ( (tval0 > 0) == (
(Bx-Ax)*(Cy-Ay) - (By-Ay)*(Cx-Ax) > 0) );
boolean test1 = (tval1 == 0) || ( (tval1 > 0) == (
(Cx-Bx)*(Ay-By) - (Cy-By)*(Ax-Bx) > 0) );
boolean test2 = (tval2 == 0) || ( (tval2 > 0) == (
(Ax-Cx)*(By-Cy) - (Ay-Cy)*(Bx-Cx) > 0) );
// flip to prevent tight loop of two triangles in
// degenerate triangulation
int it2 = itnum / 2;
boolean flip = ((it2 % 2) == 0);
// figure out which triangle to go to next
if (!test0 && !test1 && !test2) curtri = -1;
else if (!test0 && !test1) {
if (flip) {
int nextri = Delan.Walk[curtri][1];
if (nextri >= 0) curtri = nextri;
else curtri = Delan.Walk[curtri][0];
}
else {
int nextri = Delan.Walk[curtri][0];
if (nextri >= 0) curtri = nextri;
else curtri = Delan.Walk[curtri][1];
}
}
else if (!test1 && !test2) {
if (flip) {
int nextri = Delan.Walk[curtri][2];
if (nextri >= 0) curtri = nextri;
else curtri = Delan.Walk[curtri][1];
}
else {
int nextri = Delan.Walk[curtri][1];
if (nextri >= 0) curtri = nextri;
else curtri = Delan.Walk[curtri][2];
}
}
else if (!test2 && !test0) {
if (flip) {
int nextri = Delan.Walk[curtri][0];
if (nextri >= 0) curtri = nextri;
else curtri = Delan.Walk[curtri][2];
}
else {
int nextri = Delan.Walk[curtri][2];
if (nextri >= 0) curtri = nextri;
else curtri = Delan.Walk[curtri][0];
}
}
else if (!test0) curtri = Delan.Walk[curtri][0];
else if (!test1) curtri = Delan.Walk[curtri][1];
else if (!test2) curtri = Delan.Walk[curtri][2];
else foundit = true;
// Return -1 if outside of the convex hull
if (curtri < 0) foundit = true;
if (foundit) tri[i] = curtri;
}
}
return tri;
}
/** convert an array of values in R^DomainDimension to an array of 1-D
indices */
public int[] valueToIndex(float[][] value) throws VisADException {
if (value.length < DomainDimension) {
throw new SetException("Irregular2DDSet.valueToIndex: value dimension " +
value.length + " not equal to Domain dimension " +
DomainDimension);
}
int[] tri = valueToTri(value);
int[] index = new int[tri.length];
for (int i=0; i<tri.length; i++) {
if (tri[i] < 0) {
index[i] = -1;
}
else {
// current values
float x = value[0][i];
float y = value[1][i];
// triangle indices
int t = tri[i];
int t0 = Delan.Tri[t][0];
int t1 = Delan.Tri[t][1];
int t2 = Delan.Tri[t][2];
// partial distances
float D00 = Samples[0][t0] - x;
float D01 = Samples[1][t0] - y;
float D10 = Samples[0][t1] - x;
float D11 = Samples[1][t1] - y;
float D20 = Samples[0][t2] - x;
float D21 = Samples[1][t2] - y;
// distances squared
float Dsq0 = D00*D00 + D01*D01;
float Dsq1 = D10*D10 + D11*D11;
float Dsq2 = D20*D20 + D21*D21;
// find the minimum distance
float min = Math.min(Dsq0, Dsq1);
min = Math.min(min, Dsq2);
if (min == Dsq0) index[i] = t0;
else if (min == Dsq1) index[i] = t1;
else index[i] = t2;
}
}
return index;
}
/** for each of an array of values in R^DomainDimension, compute an array
of 1-D indices and an array of weights, to be used for interpolation;
indices[i] and weights[i] are null if no interpolation is possible */
public void valueToInterp(float[][] value, int[][] indices,
float[][] weights) throws VisADException {
if (value.length < DomainDimension) {
throw new SetException("Irregular2DDSet.valueToInterp: value dimension " +
value.length + " not equal to Domain dimension " +
DomainDimension);
}
int length = value[0].length; // number of values
if ( (indices.length < length) || (weights.length < length) ) {
throw new SetException(
"Irregular2DSet.valueToInterp: lengths don't match");
}
int[] tri = valueToTri(value);
for (int i=0; i<tri.length; i++) {
if (tri[i] < 0) {
indices[i] = null;
weights[i] = null;
}
else {
// indices and weights sub-arrays
int[] ival = new int[3];
float[] wval = new float[3];
// current values
float x = value[0][i];
float y = value[1][i];
// triangle indices
int t = tri[i];
int t0 = Delan.Tri[t][0];
int t1 = Delan.Tri[t][1];
int t2 = Delan.Tri[t][2];
ival[0] = t0;
ival[1] = t1;
ival[2] = t2;
// triangle vertices
float x0 = Samples[0][t0];
float y0 = Samples[1][t0];
float x1 = Samples[0][t1];
float y1 = Samples[1][t1];
float x2 = Samples[0][t2];
float y2 = Samples[1][t2];
// perpendicular lines
float C0x = y2-y1;
float C0y = x1-x2;
float C1x = y2-y0;
float C1y = x0-x2;
float C2x = y1-y0;
float C2y = x0-x1;
// weights
wval[0] = ( ( (x - x1)*C0x) + ( (y - y1)*C0y) )
/ ( ((x0 - x1)*C0x) + ((y0 - y1)*C0y) );
wval[1] = ( ( (x - x0)*C1x) + ( (y - y0)*C1y) )
/ ( ((x1 - x0)*C1x) + ((y1 - y0)*C1y) );
wval[2] = ( ( (x - x0)*C2x) + ( (y - y0)*C2y) )
/ ( ((x2 - x0)*C2x) + ((y2 - y0)*C2y) );
// fill in arrays
indices[i] = ival;
weights[i] = wval;
}
}
}
public Object cloneButType(MathType type) throws VisADException {
if (ManifoldDimension == 1) {
return new Irregular2DSet(type, Samples, newToOld, oldToNew,
DomainCoordinateSystem, SetUnits, SetErrors);
}
else {
return new Irregular2DSet(type, Samples, DomainCoordinateSystem,
SetUnits, SetErrors, Delan);
}
}
/* run 'java visad.Irregular2DSet' to test the Irregular2DSet class */
public static void main(String[] argv) throws VisADException {
float[][] samp = { {139, 357, 416, 276, 495, 395, 578, 199},
{102, 44, 306, 174, 108, 460, 333, 351} };
RealType test1 = RealType.getRealType("x");
RealType test2 = RealType.getRealType("y");
RealType[] t_array = {test1, test2};
RealTupleType t_tuple = new RealTupleType(t_array);
Irregular2DSet iSet2D = new Irregular2DSet(t_tuple, samp);
// print out Samples information
System.out.println("Samples:");
for (int i=0; i<iSet2D.Samples[0].length; i++) {
System.out.println("#"+i+":\t"+iSet2D.Samples[0][i]
+", "+iSet2D.Samples[1][i]);
}
System.out.println();
// test valueToIndex function
System.out.println("valueToIndex test:");
float[][] value = { {164, 287, 311, 417, 522, 366, 445},
{131, 323, 90, 264, 294, 421, 91} };
int[] index = iSet2D.valueToIndex(value);
for (int i=0; i<index.length; i++) {
System.out.println(value[0][i]+", "
+value[1][i]+"\t--> #"+index[i]);
}
System.out.println();
// test valueToInterp function
System.out.println("valueToInterp test:");
int[][] indices = new int[value[0].length][];
float[][] weights = new float[value[0].length][];
iSet2D.valueToInterp(value, indices, weights);
for (int i=0; i<value[0].length; i++) {
System.out.println(value[0][i]+", "+value[1][i]+"\t--> ["
+indices[i][0]+", "
+indices[i][1]+", "
+indices[i][2]+"]\tweight total: "
+(weights[i][0]+weights[i][1]+weights[i][2]));
}
System.out.println();
}
/* Here's the output:
iris 136% java visad.Irregular2DSet
Samples:
#0: 139.0, 102.0
#1: 357.0, 44.0
#2: 416.0, 306.0
#3: 276.0, 174.0
#4: 495.0, 108.0
#5: 395.0, 460.0
#6: 578.0, 333.0
#7: 199.0, 351.0
valueToIndex test:
164.0, 131.0 --> #0
287.0, 323.0 --> #7
311.0, 90.0 --> #1
417.0, 264.0 --> #2
522.0, 294.0 --> #6
366.0, 421.0 --> #5
445.0, 91.0 --> #4
valueToInterp test:
164.0, 131.0 --> [0, 3, 7] weight total: 0.99999994
287.0, 323.0 --> [2, 3, 7] weight total: 1.0
311.0, 90.0 --> [0, 1, 3] weight total: 1.0
417.0, 264.0 --> [2, 3, 4] weight total: 1.0
522.0, 294.0 --> [2, 4, 6] weight total: 1.0
366.0, 421.0 --> [2, 5, 7] weight total: 1.0
445.0, 91.0 --> [1, 3, 4] weight total: 1.0
iris 137%
*/
}
visad archives: