Helmut Simonis and Barry O'Sullivan - Almost Square Packing

Square Packing Almost Square Packing
Almost Square Packing

Helmut Simonis and Barry OSullivan
Cork Constraint Computation Centre Computer Science Department University College Cork Ireland
CSCLP 2010
Points to Remember
Almost Square Packing: Generalization of Square Packing (Simonis and OSullivan, CP 2008) Allows rotation of items, extra degree of freedom Interleave orientation of item with interval choice for X Redundant constraints similar to square packing Further decomposition possible, treat X variables on their own Does not need disjoint2 and second cumulative constraint Problems up to N=26 solved optimally (very similar to square packing)
Consider almost squares, rectangles n (n + 1) Pack all items 1 2, 2 3, ..., n (n + 1) Into smallest possible rectangle Items are non-overlapping
Background
Suggested by Prof. Des MacHale, Math, UCC Solved packing up to 13 14 by hand Starting point for Barrys and my interest in packing My objection: This is too complicated
Look at square packing Known results, benchmark comparison
Simpler Problem: Square Packing
Consider squares, rectangles n n Pack all items 1 1, 2 2, ..., n n Into smallest possible rectangle Items are non-overlapping
Outline
Problem (N=26)
10 15 43 7 6 8 9 16 2 12 20 11 5 26 21 23 14 13 25 17 19
24
22
18
Problem Decomposition
Search for candidate enclosing rectangle Area must be larger than sum of items to be placed Search in order of increasing area
and increasing squareness
Check each candidate for (in)feasibility until rst solution is found Observation: Only limited number of candidates explored
Basic Model
Width
W H X,Y Height
disjoint2 cumulative
cumulative
Interval Based Strategies
Key Idea: Fixing intervals, not values Fixing variables to values is too restrictive Select area in which item is placed Allows items to shift slightly Restrict domain to intervals Only at end x actual values
10
Forcing Compulsory Parts

Small Interval
Xl Xr
Number of intervals: large
Compulsory parts
Xl
Xr
Large Interval Number of intervals: small
11
Model Improvement: Dominance Criterion

X2 X1 W2 B X2 , Y2 D W1 Y2 + H2 Y1 + H1 W2 H1 D X2 , Y2 Y2 Y1 A H2 H2
X2 + W2 X1 + W1
X1 , Y1
12
Strategies Comparison
1e+06
100000
10000
1000 Time[s]
100 Korf BlueBlocker naive naive gaps xtheny disj semantic4 semantic dual interval split xy 16 18 20 22 Problem Size 24 26 28
10
0.1
0.01
13
Optimal Solution (N=27, CP 2008)
17
15
18
12
21
23
22
10
16
11
5 3 27
26
24
25
14
Even better results by R. Korf in IJCAI 2009
13
19
20
14
Outline
15
Back to the Future
Apply lessons learned to almost square packing Added degree of freedom, rotation of items Weak impact, length only changes by one But 2n additional choices
16
Almost Square, Optimal Solution (N=26)
14x13
12x13
15x14
20x21
23x22
23x24
18x17
3x4 4x5
6x7
5x6
17x16
10x11 11x12
7x8
2x3 2x1
22x21
19x18
19x20
10x9
24x25
25x26
27x26
9x8
15x16
17
Visualization: Search Tree (N=20)

w20
2
20
x202
1..7
w192
19
x196
1..6
w182
18
2 6 x187 1..6 8..13 17..22
77387 66481
43 27
w172
17 18 55775 46472
x178
1..6
w16
16 17 6077 4580
x16
10
1..6
w15
15 16 1297 919
x15
20
w14
14
15
543 416
x14
1..5
13 1
21
2
157 87
w13 x13
13
16
1..4
w12
12
2 x123 19..22 23..26
10 8 13 7
w112
11 12
1 4 x118 14..17 18..21 24
141 76
5 3
w102
10
x1010
14..17
w9 2
9
x9 w8
8
19..21 1
x8 w7
7
28
2
x7 w6
6
28
2
x6 w5
5
21
1
x5 w4
4
19
1
x4
20
w3 2
3 4
x3
14
w2 2
2 3
x2
21
1 xx20
1
1 xx19
xx18
18
xx17
1
xx16
1
xx15
20
xx14
21
xx13
1
xx12
24
xx11
24
xx10
14
xx91
19
xx81
28
xx71
28
xx61
21
xx51
19
xx41
20
xx31
14
xx2 y20
41 25 4 3 40 23
21
20 27 37 38 25..31
1..7 8..14 18..24
y19
1 2 1 1 13 5
19 57 58 1 7
1..7 8..14 46..52 58..64 69
y18 y17 y16

1
44
1..6 1
52
1
y15
1 4 9
62..66 67..71 74
y14 y13 y12
15
31..35 1
17
27
20..23 1 y1117 47..50 63
y101
20
y9 y8 y7 y6 y5 y4 y3 y2
53
8
47..49 1
56
1
47
1
63
1
69
1
17
1
31
yy20
31
yy19
69
yy18
1
yy17
52
yy16
1
yy15
74
1 yy14
33
1 yy13
17
1 yy12
20
1 yy11
63
1 yy10
20
yy91
53
yy8 yy7 yy6 yy5 yy4 yy3 yy2
47
1
56
1
47
1
63
1
69
1
17
1
31
18
Problem with Tree View
Showing complete tree is clearly infeasible We only show path to solution But most time is spent in non-solution parts of tree
19
Search Choices
Based on best method for square packing Assign X intervals, x X values, assign Y intervals, x Y values When to x orientation? Eager Before assigning X intervals Lazy After assigning X intervals Interleaved Mixed with X interval assignment
20
Eager Orientation (N=17)

1e+07 FAIL TRY 1e+06
100000
10000 Nodes 1000
100
10
1 0 10 20 30 40 Depth 50 60 70 80
21
Lazy Orientation (N=17)

1e+07 FAIL TRY 1e+06
100000
10000 Nodes 1000
100
10
1 0 10 20 30 40 Depth 50 60 70 80
22
Interleaved Orientation (N=17)

100000 FAIL TRY
10000
1000 Nodes
100
10
1 0 10 20 30 40 Depth 50 60 70 80
23
Comparison (N=17)
1e+07 Eager Lazy Interleaved 1e+06
100000
10000 Nodes 1000
100
10
1 0 10 20 30 40 Depth 50 60 70 80
24
Node Distribution (N=20)

20_35_88 100000 W and X Interval X Assignment Y Interval Y Assignment FAIL TRY
10000
1000 Nodes
100
10
1 0 10 20 30 40 50 Depth 60 70 80 90 100
25
Problem with Slack (N=21)

21_46_77 1e+06 "flat_21_46_77.dat"
100000
10000
Nodes
1000
100
10
1 0 10 20 30 Depth 40 50 60 70
26
Impact of Interval Size (N=26)

26_42_156 1e+07 0.25 0.30 0.35 0.40 0.45
1e+06
100000
10000 Nodes 1000
100
10
1 0 20 40 60 Depth 80 100 120 140
27
Basic Model Results (SICStus Prolog 4.04), 2008 Desktop

N 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 Loss 0.00 0.00 1.79 0.00 0.00 1.82 0.45 0.00 0.96 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.20 0.00 0.00 0.00 0.00 0.00 First Solution Back Time 2 00:00 4 00:00 16 00:00 19 00:00 6 00:00 54 00:00 323 00:00 99 00:00 546 00:00 1900 00:00 2937 00:00 14440 00:00 15967 00:01 210878 00:14 9734 00:00 102235 00:08 351659 00:34 14036353 21:38 58206362 01:37:30 14490682 30:12 27475258 55:05 35282646 01:23:12 92228265 03:28:20 Sols 8 16 216 65 12 9170 1854 4 25180 42 4 4 544 16 110288 526 1988 3250117 688 6784 96 1007780 1056 All Solutions Back 6 14 24 76 83 3137 1379 268 13795 6197 9604 50592 48711 398759 152032 3240741 3612859 720146935 122563947 136039535 99731414 Time 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:02 00:00 00:00 00:03 00:03 00:27 00:24 04:26 05:52 25:13:20 03:23:19 04:38:40 03:20:37
28
Redundant Constraints
Forbidden gaps Ignore small blocks Symmetries of aligned blocks(?)
29
Forbidden Gaps
Large items can be placed either directly on border or away from border Otherwise gap must be lled with small items and slack For each gap size, pre-compute length of item that can not be covered Slight complication due to rotation of items, use constructive disjunction
Remove most forbidden gaps at setup by reducing domains Further reduction when item is oriented
Does not handle interaction of multiple gaps
30
Pseudo Solution Removing 1 2 Block
14x13 6x5 12x11 7x6 13x12 4x5 15x14 7x8 10x11 4x3 2x3 16x17 16x15
10x9 8x9
31
Redundant Constraint Result

N 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 Basic Model Back Time 2 00:00 4 00:00 16 00:00 19 00:00 6 00:00 54 00:00 323 00:00 99 00:00 546 00:00 1900 00:00 2937 00:00 14440 00:00 15967 00:01 210878 00:14 9734 00:00 102235 00:08 351659 00:34 14036353 21:38 58206362 01:37:30 14490682 30:12 27475258 55:05 35282646 01:23:12 92228265 03:28:20 Not One Back Time 2 00:00 3 00:00 16 00:00 18 00:00 5 00:00 54 00:00 323 00:00 99 00:00 546 00:00 1900 00:00 2936 00:00 14425 00:00 9338 00:00 210850 00:13 9734 00:00 102235 00:08 355964 00:33 10859861 16:01 58214183 01:33:03 14490682 29:16 27475258 53:11 35502799 01:21:25 92228259 03:22:33 Gap Back 2 2 6 10 17 27 159 54 274 1040 1505 7632 7264 107639 5550 13694 157312 9499957 17312971 6400629 9801577 13030167 29432477 Time 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:07 00:00 00:01 00:14 14:14 24:37 11:01 16:39 25:16 55:38 Both Back 2 1 6 9 10 27 159 54 274 1040 1501 7617 3989 107611 5550 13690 161410 6524396 17319946 6400629 9801577 13232221 29432467 Time 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:00 00:07 00:00 00:01 00:14 09:13 23:54 10:33 16:10 25:15 54:08
32
Impact of Interval Size

12000 10000 8000 Time[s] 6000 4000 2000 0 0.1
17 18 19 20 21 22 23 24 25 26
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Parameter
33
Further Problem Decomposition
Do we really need the full basic model? Most search in X variable search phase Postpone setting up Y variables, disjoint2 and 2nd cumulative Only create them when complete solution for X variables found Uses just single cumulative constraint We can even replace disjoint2 by reied inequalities
34
Decomposition Results
N 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 Basic Model Back Time 2 00:00 1 00:00 6 00:00 9 00:00 10 00:00 27 00:00 159 00:00 54 00:00 274 00:00 1040 00:00 1501 00:00 7617 00:00 3989 00:00 107611 00:07 5550 00:00 13690 00:01 161410 00:14 6524396 09:13 17319946 23:54 6400629 10:33 9801577 16:10 13232221 25:15 29432467 54:08 With Redundant Constraints Decomposed Model Decomposed Reied Back Time Back Time 2 00:00 2 00:00 1 00:00 1 00:00 6 00:00 6 00:00 9 00:00 9 00:00 10 00:00 10 00:00 27 00:00 27 00:00 176 00:00 176 00:00 54 00:00 54 00:00 301 00:00 301 00:00 1040 00:00 1040 00:00 1501 00:00 1501 00:00 7617 00:00 7617 00:00 3989 00:00 3989 00:00 107611 00:05 107611 00:06 5550 00:00 5550 00:00 13690 00:00 13690 00:00 161410 00:11 161410 00:16 6524396 07:10 6524396 08:05 17319946 19:13 17319946 21:50 6400629 08:04 6400629 08:58 9801577 12:11 9801577 13:07 13232221 20:07 13232773 23:34 29432467 40:26 29432467 43:51
35
Comparison with (Korf et al., 2010)
Size 17 18 19 20 21 22 23 24 25 26
Area 3457 3076 3576 3588 3991 4492 40115 40130 45130 42156
Korf, Moftt and Pollack Nodes Times 6,889,973 :07 22,393,428 :26 11,918,834 :11 608,635,198 12:50 792,197,287 23:21 4,544,585,807 1:49:32 32,222,677,089 15:06:56 41,976,042,836 18:39:34 557,540,262,189 12:11:30:32 -
Redundant Back Times 107611 00:07 5550 00:00 13690 00:01 161410 00:14 6524396 09:13 17319946 23:54 6400629 10:33 9801577 16:10 13232221 25:15 29432467 54:08
Decomposition Back Times 107611 00:05 5550 00:00 13690 00:00 161410 00:11 6524396 07:10 17319946 19:13 6400629 08:04 9801577 12:11 13232221 20:07 29432467 40:26
36
Future Work
SICStus Prolog cumulative does not know about volume of items Domain variables [n, n + 1] for duration and resource use Does not know that w h is n (n + 1) Underestimates volume as n n Improved energetic reasoning possible (CHIP)
37
Points to Remember
Almost Square Packing: Generalization of Square Packing (CP 2008) Interleave orientation of item with interval choice for X Further decomposition possible, treat X variables on their own Does not need disjoint2 and second cumulative constraint Problems up to N=26 solved optimally (very similar to square packing) Outperforms best published results by 103
38

Helmut Simonis and Barry O'Sullivan - Almost Square Packing

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Helmut Simonis and Barry O'Sullivan - Almost Square Packing

Uploaded by

Copyright:

Available Formats

Square Packing Almost Square Packing

Almost Square Packing