Still trying to figure this out

code/nnv/engine/set/Star.m

@@ -1166,9 +1166,21 @@
             if obj.dim ~= S.dim
                 error('Two sets have different dimension');
             end
+
+            if isa(obj.V, 'single') || isa(obj.C,'single') || isa(obj.d, 'single')
+                S1 = obj.changeVarsPrecision('double');
+            else
+                S1 = obj;
+            end
+
+            if isa(S.V, 'single') || isa(S.C,'single') || isa(S.d, 'single')
+                S2 = S.changeVarsPrecision('double');
+            else
+                S2 = S;
+            end
 
-            P1 = obj.toPolyhedron();
-            P2 = S.toPolyhedron();
+            P1 = S1.toPolyhedron();
+            P2 = S2.toPolyhedron();
 
             bool = (P1 <= P2);
 

code/nnv/examples/NN/medmnist/process_debugNodule.m

@@ -31,3 +31,48 @@
 figure;
 Star.plot(R3,'r');
 
+
+% Let's see what the differences in the layer-by-layer sets are...
+
+% Layer 1
+r1 = R1a{5}{1};
+r2 = R2a{5}{1};
+r3 = R3a{5}{1};
+
+rs1 = r1.toStar;
+rs2 = r2.toStar;
+rs3 = r3.toStar;
+
+% The next 3 statements should return true
+% bool1 = rs1.isSubSet(rs2);
+% bool2 = rs1.isSubSet(rs3);
+% bool3 = rs2.isSubSet(rs3);
+% this takes too long as stars are high dimensional (polyhedron not very
+% efficient)
+
+% get ranges (overapprox) and see if the smaller bounds are within te
+% larger ones (rs1 < rs2 < rs3)
+[rs1L, rs1U] = rs1.estimateRanges;
+[rs2L, rs2U] = rs2.estimateRanges;
+[rs3L, rs3U] = rs3.estimateRanges;
+
+b1L = all(rs1L >= rs2L);
+b1U = all(rs1U <= rs2U);
+b2L = all(rs1L >= rs3L);
+b2U = all(rs1U <= rs3U);
+b3L = all(rs2L >= rs3L);
+b3U = all(rs2U <= rs3U);
+
+% Lower bounds seem fine, but ot the upper bounds...
+
+% where do these assertions fail?
+u1 = find(rs1U > rs2U);
+u2 = find(rs1U > rs3U);
+u3 = find(rs2U > rs3U);
+
+% these fail on all the perturbed pixels...
+uu11 = rs1L(u1);
+uu12 = rs2L(u1);
+
+% we are not defining the constraints on the input set correctly, some
+% pixel values are outside the valid ranges...