software Questions - OR-Exchange

Performance of LP solvers

2010-07-10T17:59:41Z

Hi Guys, I am working on a LP relaxation of a large scale industrial problem ( over half a million variables and several tenths of thousands constraints ), using coin-or clp. Do you think that it is worth investing time in tests with a different open source solver such as glpk or maybe commercial cplex or express-mp ? I know that the question is very general, more specifically I might ask if you know of cases when commercial solvers can be say order of magnitude faster when solving pure linear problems ? and as for open source solvers can such great disproportion in efficiency be observed in some cases ?

Best regards Luke

Ideas for OR/analytics iPad app

2010-06-16T14:56:00Z

I've recently got my hands on an iPad, and I'm eager to turn my hand to creating some apps. It is a great interface for many tasks, I'm wondering if it would be suitable for an analytics interface... I'd love some ideas and feedback.

The first idea I had was for a predictive modelling interface, where models could be created for data that is stored on a server somewhere, possibly in the cloud. What would be the minimum feature set for such an app to be useful?

Another idea could be for a discrete-event simulation environment...

Any other ideas? Are there any OR-specific applications that would suit this interface?

Market basket anlaysis with two types of products with R

2010-07-09T17:49:29Z

I have two types of products, videos and sounds, and I want to understand which sounds are purchased often with which videos or sets of videos. Pretty much a market basket analysis but just one in which the lhs of the rules is only video products and the rhs only audio products.

I'm using R and the package arules doesn't seem to have a good option to do this, does anyone know how to do this with R or one of its libraries. Else I'll have to mine simpler rules which would be a shame.

Thanks

Ideas for the INFORMS Data Mining Contest

2010-06-27T02:22:52Z

INFORMS has recently announced its data mining competition. They have posted the famous billion Dollar question of "how to do day trading" and whoever solves the challenge may get some recognition (and no money from INFORMS!).

I am wondering, what are the interesting models, resources, software to tackle this question? I have used R and its GLM package to perform logistic regressions. The following is my code for getting the AUC of 0.611879 (I am currently in the 3rd place) feel free to improve upon this code (additionally, any citation to this code or anything that can help me get a faculty position in a year is hugely appreciated)

R code (I ran it on an Ubuntu Linux): please note that I developed this piece of code for personal use I noticed there are wording and grammatical errors in the comments.

# Logit Regression Model
rm(list = ls(all = TRUE))
setwd('~/Desktop/Informs\ Datamining\ Contest/Data/')
training.data <- read.csv("TrainingData.csv", header = TRUE) 

head(training.data)

plot(training.data$Timestamp,training.data$Variable141LAST_PRICE, type='l')
attach(training.data)

#Finding the maximum positive and negative correlation
names(training.data)
corrs <- as.null()

for (stock.name in names(training.data)) {
    correlation <- cor(eval(as.name(eval(stock.name))),TargetVariable)
    corrs <- rbind(corrs, correlation)
    print(paste("Correlation Between TargetStock and", stock.name," = ", correlation))
}
correls <-data.frame(ComparedStock = names(training.data), Correlations = corrs)
correls[order(correls$Correlations),] 

write.table(correls[order(correls$Correlations),],file="Correlations.csv",sep=",",
            quote=FALSE)

# Prediction Model
# Logistic Regression Model

mylogit<- glm(TargetVariable~Variable101OPEN+Variable101LOW+Variable101HIGH+Variable101LAST_PRICE+Variable133LOW+Variable133OPEN+Variable78LOW, family=binomial(link="logit"), na.action=na.pass)

#Readign Result Data
result.data <- read.csv("ResultData.csv", header = TRUE) 
head(result.data)
FinalPrediction<-predict(mylogit,newdata=result.data,type="response")

#Saving Data
template.data <- read.csv("template.csv", header = TRUE) 
my.output<-data.frame(template.data$Timestamp,FinalPrediction)
head(my.output)
names(my.output) <-c("Timestamp","TargetVariable")

write.table(my.output,file="submit.csv",sep=",",row.names=FALSE,
            quote=FALSE)

My email address is linux_jvm@yahoo.com if you have any questions.

Failure of OR-Exchange?

2010-04-14T02:19:06Z

Stack Exchange is changing how they will work. Instead of having individuals come up with sites, they have a philosophy of how sites "should" work. It appears that OR-Exchange does not match with their philosophy. Among other problems, we had a poor April 8 (as I understand it), so are now on a three month time limit.

Can anyone actually understand http://blog.stackexchange.com/post/518474918/stack-exchange-2-0 and see whether OR-Exchange can continue? I have tried to go through it a few times, but the gap between their world view and mine seems far too wide.

Of course, it is their software and they can do as they wish. But suggestions on how we can either work in the proposed structure or alternative software to provide the same services are very much appreciated.

What is your software optimization solver of preference?

2010-04-15T12:55:39Z

Please post an answer that is your favorite solver. If your favorite solver is already listed vote it up. Examples include CPLEX, Gurobi, GLPK, Symphony, lpsolve, Excel Solver, etc...

What file format for problem definition is suitable for OR-Exchange?

2010-04-16T12:03:49Z

What would be a text format to state a mathematical programming problem that could be published here on OR-Exchange?

Especially for linear programming, I like lpsolve .lp format. It is far more readable than, for instance, the MPS format (see here for a comparison).

What other formats are out there, both human readable and widely supported?

(I would like to tag this question as "documentation" and/or "format" but I have not the right to tag currently).

Open source/free MIP Solver

2009-11-18T08:20:00Z

I am solving big optimization models (more than 100k variables) with the academic version of CPLEX but I'd like to see if I can find an open source alternative, I solve mixed integer problems (MILP) and CPLEX works great but it is very expensive if I graduate (which will hopefully happen someday in the future). so I really need to find an alternative or start writing my own optimization library (which will be painful)

Any suggestion/insight would be much appreciated

Any feedback on GLPK or lp_solve?

Extracting Gomory cuts out of Cgl (Coin-or)

2010-01-05T02:19:58Z

Hey,

I'm trying to extract Cgl Gomory cuts out of the Cgl (Cut Generation Library) of Coin-Or The following is the code I'm using to extract the cuts -

OsiCuts cutlist;
CglGomory * gomory = new CglGomory();
gomory->setLimit(100);
gomory->generateCuts(*sym, cutlist) ;

where sym is an instance of OsiSymSolverInterface (the OsiSolverInterface for Symphony). Unfortunately the code is segfaulting at generateCuts somewhere inside the method as far as I've been able to determine using gdb.

Extraction of CglProbing cuts is likewise segfaulting again inside the generateCuts method of the CglProbing class.

All other cuts seem to be working fine.

If someone could shed some light on this or even better, post/link to an example file using these cuts or a tutorial of some sort, that would great. If there's an example/tutorial for extracting cuts out of some other solver like SCIP instead of Coin-OR, that would work too.

Thanks