Embedding Formulation #73

jac0320 · 2017-11-22T19:05:00Z

@kaarthiksundar

File addition:

Added new file embedding.jl for handling formulation
separating out bin.jl

Data structure changes:

Added embedding, embedding_encode, embedding_ibs, embedding_link for options with embedding formulations

Algorithm changes and Function additions:

Add logs associated with embedding formulation
Let multi.jl fit with embedding formulations, i.e., divert from sos-2 formulation if embedding is true
Update function build_constr_block()
Introducing Binary Product MultiLinear (bpml) for more complex multilinear terms that ha binary variables in them
separating several term storage by using the modularized function
Some refactoring in operators.jl

Error schemes and default changes:

Test changes:

Added tests for bpml terms
Added tests for embedding
Cleaned up several test instances code

---------- OLD Comments -------------
@harshangrjn Please see this PR for some consideration on what should be left out.

This is the all the development for POD.jl on the embedding formulation side #64 .
It contains three main aspects,

Basic Embedding Formulation (Gray code Encoding)
Independent Branching Scheme for Compact Embedding
Binary Variable - Continuous Variable Bound restriction constraints reconstruction with embedding

The options introduced in this PR here are:

embedding::Bool -> Control the to use embedding formulation or not
embedding_encode::Any -> Direct which encoding will be used (compatibility will be checked)
embedding_ibs::Bool -> Whether to exploit the compactness of the formulation or not
embedding_link::Bool -> Add bound restriction constraints to regulate variable bounds based on active partition

Test cases are added to verify the encoding mapping and algorithm. Note that embedding formulation, regardless what additional options to choose, should yield the same results as SOS-2 formulation.

!!! This PR branch off #72.

…edding

codecov · 2017-11-22T21:14:23Z

Codecov Report

Merging #73 into master will increase coverage by 0.79%.
The diff coverage is 95.41%.

@@           Coverage Diff            @@
##           master    #73      +/-   ##
========================================
+ Coverage   78.21%    79%   +0.79%     
========================================
  Files          27     28       +1     
  Lines        5696   5898     +202     
========================================
+ Hits         4455   4660     +205     
+ Misses       1241   1238       -3

Impacted Files	Coverage Δ
src/POD.jl	`100% <ø> (ø)`	⬆️
src/utility.jl	`78.46% <ø> (ø)`	⬆️
src/algorithm.jl	`85.55% <100%> (+0.16%)`	⬆️
src/solver.jl	`61.02% <100%> (+0.87%)`	⬆️
src/log.jl	`75.45% <100%> (+0.92%)`	⬆️
test/algorithm.jl	`100% <100%> (ø)`	⬆️
test/solver.jl	`100% <100%> (ø)`	⬆️
src/multi.jl	`74.77% <89.36%> (+6.02%)`	⬆️
src/embedding.jl	`92.94% <92.94%> (ø)`

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 64539d9...a40620f. Read the comment docs.

…edding

kaarthiksundar · 2018-06-16T19:38:57Z

src/algorithm.jl

+    embedding::Bool
+    embedding_encode::Any                                       # Encoding method used for embedding
+    embedding_ibs::Bool                                         # Enable independent branching scheme
+    embedding_link::Bool                                         # Linking constraints between x and α, type 1 usse hierarchical and type 2 with big-m


What is this embedding_link for?

This is for an experiment to try out the alternative bounding constraints (y1b0+y1b1<=x<=y1b1+y2b2) when you have log # of binary variables.

Was this experiment successful - did it improve computation time and bounds?

The short answer is: it can be good. The results are not consistent for two reasons: 1) in some cases, this can help, 2)the test scope is small. I suggest we keep the current code for later experiments.
For example, with the multiKND problem,

Another example with nlp3:

kaarthiksundar · 2018-06-16T19:41:51Z

src/embedding.jl

+   return v
+end
+
+# # Version 1 implementation #


Can you clean up the commented part - or may be remove it it we are not going to use it any more. Or if you dont want to remove it - make it as a function and add a comment saying old version

This is taken care of in the most recent commit.

kaarthiksundar · 2018-06-16T19:42:31Z

src/embedding.jl

@@ -0,0 +1,253 @@
+# Make sure two things :: correct setup & compatible mapping function


Please add a comment for each function

This is taken care of in the most recent commit.

kaarthiksundar · 2018-06-16T19:43:15Z

src/multi.jl

@@ -170,8 +179,7 @@ function populate_convhull_extreme_values(m::PODNonlinearModel, discretization::
 end

 """
-    TODO: docstring
-    method for general multilinear term
+    Extreme value for gneral multilinear terms Efficient implementation


This is taken care of in the most recent commit.

kaarthiksundar · 2018-06-16T20:26:37Z

src/multi.jl

@@ -197,7 +205,7 @@ function populate_convhull_extreme_values(m::PODNonlinearModel, discretization::
 end

 """
-    Less memeory & time efficient, but a easier implementation
+    Less memeory & time efficiency, but a easier implementation


This is taken care of in the most recent commit.

kaarthiksundar · 2018-06-16T20:27:06Z

src/multi.jl

@@ -197,7 +205,7 @@ function populate_convhull_extreme_values(m::PODNonlinearModel, discretization::
 end

 """
-    Less memeory & time efficient, but a easier implementation
+    Less memeory & time efficiency, but a easier implementation
 """
 function _populate_convhull_extreme_values(discretization::Dict, ml_indices::Any, λ::Dict, extreme_point_cnt::Int)


make function names consistent why is there an underscore before the function name

This function _populate_convhull_extreme_values() is not actively used. But it serves the purpose of verifying the populate_convhull_extreme_values() just in case the multi-dimensional scheme in Julia is different.

kaarthiksundar · 2018-06-16T20:28:09Z

src/multi.jl

@@ -347,18 +381,16 @@ function valid_inequalities(m::PODNonlinearModel, discretization::Dict, λ::Dict
        return
    end

+    # Minimum Formulation with Boundary Cuts


do we still need this - let me discuss with harsha and russell and get back to you on this - for now have it

Okay. This is good for testing purposes.

You can remove this function with its solver options.

Okay. This is addressed in the most recent commit. Just a note, this option is removed in a later PR. We can do it here and I will rebase the other branches.

kaarthiksundar

Please address the comments and get back

kaarthiksundar

Please address the comments

kaarthiksundar

The one thing I am not completely satisfied about is the naming convention for the embedding options. I will try to come up with more meaningful names and we can do another update later.

jac0320 · 2018-06-18T19:27:56Z

The "embedding" will be later refactored as "ebd". The rest of the string will be the same. This is the final stage of such options.

Site Wang and others added 22 commits September 22, 2017 08:56

Refactoring some basics

3f4c69e

basic layout and sos1 compatability

a4b642e

partial implementation for sos2 case

2226e23

Mapping for SOS-2 embedding and formulation implementation

25122b0

Bug fixed

9b179d7

Some minor fixes. Still need more...

74d1627

Logging and clean up; new cuts tested

644cf67

Sync for local work

cb1e4fd

Partial implementation for transport, too tired to continue

6f8e1af

Cleaning out useless functions, keeping the essentials

1e5d756

Independent branching scheme added for embedding, working well!

179cf9a

Bug fix on monomial terms

4759f19

Minor clean up on the embedding formulation

eac055a

Merge branch 'master' of https://github.com/lanl-ansi/POD into embedding

671ffb2

Linking variable method A implemented

ed2f768

Tested linkin constraints method A

47cd560

Merge branch 'embedding' of https://github.com/lanl-ansi/POD into emb…

3d81636

…edding

New linking constraints version 1

a64e396

All versions of the embedding linking is there. Start testing

e8b1bf5

Merge branch 'master' of https://github.com/lanl-ansi/POD into embedding

ec0f2dd

Resolved conflicts

63dbe3b

Ready for large scale testing

8885d6b

jac0320 requested a review from harshangrjn November 22, 2017 19:05

jac0320 closed this Nov 22, 2017

Site Wang added 2 commits November 22, 2017 12:29

Intermedia commit

883f413

Test cases added for this branch

d73c2b2

jac0320 reopened this Nov 22, 2017

jac0320 mentioned this pull request Nov 23, 2017

Documentation and Logs #54

Open

jac0320 closed this Feb 21, 2018

jac0320 added 2 commits June 11, 2018 15:32

Merge branch 'master' of https://github.com/lanl-ansi/POD.jl into emb…

4184ec9

…edding

Ready for review

e1e5942

jac0320 reopened this Jun 11, 2018

Fixed file include

caa72eb

jac0320 changed the title ~~Embedding Formulation~~ Embedding Formulation and Strengthening BPML terms Jun 11, 2018

jac0320 requested review from kaarthiksundar and removed request for harshangrjn June 11, 2018 21:58

jac0320 changed the title ~~Embedding Formulation and Strengthening BPML terms~~ Embedding Formulation Jun 14, 2018

kaarthiksundar reviewed Jun 16, 2018

View reviewed changes

kaarthiksundar requested changes Jun 16, 2018

View reviewed changes

Added a few docstring and amended some typos in exisiting ones

01b6d83

kaarthiksundar requested changes Jun 18, 2018

View reviewed changes

Removing option convhull_formulation_minib

a40620f

kaarthiksundar approved these changes Jun 18, 2018

View reviewed changes

kaarthiksundar merged commit 9ff7bf1 into master Jun 18, 2018

jac0320 deleted the embedding branch June 18, 2018 19:18

jac0320 mentioned this pull request Jun 21, 2018

Embedding Formulation and Redundancy #64

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Embedding Formulation #73

Embedding Formulation #73

jac0320 commented Nov 22, 2017 •

edited

Loading

codecov bot commented Nov 22, 2017 •

edited

Loading

kaarthiksundar Jun 16, 2018

jac0320 Jun 18, 2018

kaarthiksundar Jun 18, 2018

jac0320 Jun 18, 2018

kaarthiksundar Jun 16, 2018

jac0320 Jun 18, 2018

kaarthiksundar Jun 16, 2018

jac0320 Jun 18, 2018

kaarthiksundar Jun 16, 2018

jac0320 Jun 18, 2018

kaarthiksundar Jun 16, 2018

jac0320 Jun 18, 2018

kaarthiksundar Jun 16, 2018

jac0320 Jun 18, 2018

kaarthiksundar Jun 16, 2018

jac0320 Jun 18, 2018

kaarthiksundar Jun 18, 2018

jac0320 Jun 18, 2018

kaarthiksundar left a comment

kaarthiksundar left a comment

kaarthiksundar left a comment

jac0320 commented Jun 18, 2018

		@@ -0,0 +1,253 @@
		# Make sure two things :: correct setup & compatible mapping function

Embedding Formulation #73

Embedding Formulation #73

Conversation

jac0320 commented Nov 22, 2017 • edited Loading

codecov bot commented Nov 22, 2017 • edited Loading

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kaarthiksundar left a comment

Choose a reason for hiding this comment

kaarthiksundar left a comment

Choose a reason for hiding this comment

kaarthiksundar left a comment

Choose a reason for hiding this comment

jac0320 commented Jun 18, 2018

jac0320 commented Nov 22, 2017 •

edited

Loading

codecov bot commented Nov 22, 2017 •

edited

Loading