So we're taking our shot at R*Stat and have provided it with plenty of data. Plenty of rows, plenty of columns. We've got everything set up and the data is super-clean in the input file.

We're following the steps in the Creating a Scoring Application quick-start guide from Information Builders and have worked our way up through to step 4, where the guide says to "Visualize the Decision Tree Model". When we do this, no tree. It appears that we're getting 15 pages of rules output (textual) and no graphical representation at all.

Did we just ship the thing too many data columns? Generally with R*Stat they tell you more is better, but I'm concerned we may have given it so many options that it is incapable of rendering the output tree based on the results.

Our goal is to look at a queue (i.e., people waiting in line) to see if we can determine what processes are making for long wait times. Multiple services pull from the same queue, so it could be that one or two (or more) services are causing long wait times. This appears to occur in some of our locations, but not others. It's not immediately apparent what's causing the bottleneck when inspecting the data visually.

So we've put the count of each of the services service provided, the average wait for people queueing for that service, the average service time for people queueing for that service, plus the average wait time overall, and date and location information for each line of data. The result is 48 columns of data. Have we just done blowed up R*Stat? It cannot render the decision tree image.

This message has been edited. Last edited by: FP Mod Chuck,

I'm also concerned that we may be asking a question that R*Stat is not capable of answering. It may be that we need to re-form the question. As it stands, we're looking to predict the wait time for a location on a particular day, using anticipated service counts. That is, we're asking for an analog answer, a measurement.

The example in the doc provides a binary answer -- Yes or No.

Is R*Stat capable of providing analog answers like the one we're requesting?

How many data columns? As far as I know, the limit is 200 columns, including the answer.

In our first attempt, our "training" data had 250 col of input, 5 years of data with over 83000 rows. RSTAT created the function but it never returned anything. We opened a problem report with IBI and the consultant found an obscure note somewhere that said the limit was 200 columns including the answer. We ended up with having to create 2 RSTAT modules, the first taking about 150 columns and the second taking about 50 with some being used in both. The data scientist said that around 80 of the columns had no effect.

If you are asking a YES/NO question, make sure you are asking something that is between 50/50 and 40/60. We did a weighted score (between 0 and 1) and failed miserably (as I knew we would) because we were trying to forecast an event that happens about .02 times out of a 100. RSTAT can provide a YES/NO answer or a numeric score, but the score is a forecast of an event happening, not a value. For example, it may tell you the there is a 50% chance of wait time of 30 minutes, but I don't think it will tell the expected weight time, could be wrong, but that is my limited understanding.

This message has been edited. Last edited by: jgelona,

John

I e-mailed with the RSTAT expert in NY. She requested the data to work with. I will send you a PM with her contact info.