I’ve watched it several times (you probably too), but it’s still fun:
http://boxcar2d.com/ Shows a demo of a genetic algorithm that builds and evaluates cars.
Cool stuff. Watch the color of the parts changing, and get a deeper insight on the subpage “the algorithm”.

Best,
jp – orconomy.de

Nice weather, quiet Sunday – time to solve one those logic quest Logelei of the die zeit newspaper. I wanted (sic: past) to model it with OML, and to solve it with our little OptimizationZen browser tool. It took me some time to understand that OML does not fulfill some requirements for an easy approach.

OML does not

• support set elements to be given in the model file (you may define the set itself, thou)
• know anything about sparse indices (for example the reduce model size if not all possible index combinations are valid: think of an non-complete graph)
• does not allow data to be in the model file (one always needs to have at least a second file, where the data bindings link to)

I thought, I’d better share this with you. Hopefully this saves your Sunday’s time of fruitless research.

Btw, does anyone know some good examples of data bindings to text files? Excel files are used all the way.

Best
jp

This post is about how you can work with sparse parameter sets and decision variables efficiently using MOSEL.
First let me mention that MOSEL does not allow defining sparse index sets, but there still is a way to get around this:
Consider two index sets I1 and I2 both taken from a set I = {1, 2, 3, 4} where we only want to use a few edges (i1, i2) from I1 x I2. Those are stored in a table “Edges” (all Nodes are also listed in a simple table “Nodes”):

I1	I2	Capacity
1	2	5
2	3	1
3	4	3
4	1	7

Node

1

2

3

4

So, how can you read capacity data from a Database for these edges efficiently?See the ‘dynamic’ keyword in the following code sample?. It will ensure that only the given index-tuples of I1 and I2 will be created as parameters.

model NAII
uses "mmxprs", "mmodbc"
declarations
I1: set of integer
I2: set of integer
Capacity: dynamic array(I1,I2) of real
end-declarations
SQLconnect("Driver=…")
SQLexecute("select node from nodes",I1)
SQLexecute("select node from nodes",I2)
SQLexecute("select I1, I2, Capacity from edges",Capacity)
SQLdisconnect()

Creating decision variables (mpvar) for only these edges is also pretty easy. First you need an existing array of the entries you would like to have decision variables for (in this case we reuse Capacity, otherwise setup a temp array).

declarations
Flow: dynamic array(I1,I2) of mpvar
end-declarations

You need to provide the index sets in the forall-loop in the same order as they appear in the definition of Capacity. By doing so, MOSEL may cleverly iterate thru I1 x I2 and speed up the search for existing values, especially when they are very sparse.

I’d like to thank Frank, Chris, Susanne and Oliver who work with FICO and helped a lot discussing details on MOSEL.

The code above was implemented in a tool automatically generating MOSEL Code which connects to a database driven optimization tool. The optimization tool allows users to freely create parameters and decision variables. Therefore very sparse arrays needed to be handled in an efficient way.

Ps: MPL would allow you to set the index sets as follows

INDEX
I1 := DATABASE(“Nodes”); -- Reads I1
I2 := I1; -- Same values as I1
Edges[I1,I2] := DATABASE(“Edges”); -- Read the valid edges

and read the sparse data only for valid tuples

SPARSE DATA
Capacity[I1, I2 in Edges] := DATABASE(“Edges”);

Finally define the variable only for valid tuples

Flow [I1, I2 in Edges];

—
Jens Peter Kempkes, ORCONOMY GmbH, Contact Info. Please feel free to contact me with any questions directly.