DEFISE wrote:3 steps in W5:
whip[4]: r5n4{c6 c3}- c3n9{r5 r3}- c6n9{r3 r7}- c6n8{r7 .} => -4r8c6
Singles: 4r8c4, 4r4c3, 4r5c6, 2r5c8, 2r1c7, 2r4c2, 2r6c6
whip[5]: c4n3{r4 r7}- c8n3{r7 r1}- c9n3{r1 r9}- c1n3{r9 r5}- c5n3{r5 .} => -3r4c7
Singles: 7r4c7, 7r5c5
Block/Line : 7b2r3 => -7r3c2 -7r3c3 -7r3c9
whip[2]: c1n7{r1 r9}- c9n7{r9 .} => -7r1c8
STTE
Good.
I had tried 1-step and 2-step in W8 and I inadvertently kept 8 as the max length allowed for fewer steps.
After seeing your solution, I checked what I'd get in W5. First try gave me 3 steps also:
biv-chain-cn[4]: c3n9{r5 r3} - c6n9{r3 r7} - c6n8{r7 r8} - c6n4{r8 r5} ==> r5c3≠4singles ==> r4c3=4, r5c6=4, r6c6=2, r4c2=2, r5c8=2, r1c7=2, r8c4=4
whip-cn[5]: c4n3{r4 r7} - c8n3{r7 r1} - c9n3{r3 r9} - c5n3{r9 r5} - c1n3{r5 .} ==> r4c7≠3singles ==> r4c7=7, r5c5=7
whip[1]: b2n7{r3c6 .} ==> r3c2≠7, r3c3≠7, r3c9≠7
whip-cn[4]: c4n7{r3 r7} - c8n7{r7 r1} - c9n7{r2 r9} - c1n7{r9 .} ==> r3c4≠5stte
My second step is the same as yours.
My first step is the same as what I found in W8.
I checked why the second step in W8 was a whip[8]: this long whip[8] has score 20, while the whip-cn[5] here has score 19. As a result, the method could in no case find the whip-cn[5] if length 8 was allowed.
This illustrates three things :
- it is essential to choose the maximum allowed length very carefully (my bad in my previous solution);
- steepest descent can't guarantee the best result (which is already well-known by anyone familiar with steepest descent);
- as a consequence, deviations from steepest descent (as appears to exist in my implementation of the method) may not be fundamentally bad.
Note: I also tried in W4, but the best I got after 3 tries was 5 steps.