In a decision tree algorithm, is there a known way to force a branch at a top level? For exmaple, I have 30 known decision patterns that are going to be completely different and I don't want them to intermingle. I wanted to force a branch at the top node on one of the 30 patterns so I wouldn't have to create 30 mining models per client.
Brian
Interactive training of models (including partial tree definition) is not supported in SQL Server 2005 Data Mining.
In many cases, instead of 30 models per client it is possible to create a single model containing 30 trees, by using a nested table (which, effectively, has the same effect as forcing a branch at the top). This works easily, for instance, if your target variable has 30 states.
Could you please provide a few details on your known decision patterns?
|||Bogdan, thanks for the answer!
The project essentially has two nested tables of terms (from SSIS' term extraction) that a specific item uses. Each item is in the parent case table 30 times (once per group,state or view based on your vocabulary you'd like to choose). The state (or group) is an input into the model and it does seem to work great for some of the states but not others and that's why I was trying to force it to break at the state level.
Essentially, we're trying to determine that if a record mentions term X, Y and Z that it's a great item to client. The problem is that it's a great item to show one group of clients but not others (based on their state).
Brian
|||I was able to get the model to branch by changing the scoring mechanism and by making the value that I was trying to force the break on discrete. Thanks again for your help. Is there any plans in future releases to enable users to force a break?
No comments:
Post a Comment