6
$\begingroup$

For instance, would the following question be on-topic?

In the context of training a decision tree, let $Y$ be the dependent variable and $X_1, \dots X_d$ the independent variables.

In a typical implementation of a decision tree, the conditions used to split the dependent variable only involve a single dependent variable (e.g. $X_2 > t_{X_2}$). However, some implementations use conditions (called oblique) that involve more than a single feature (e.g. $X_1+ X_2 > t_{f(X_1, X_2)}$).

In general, an oblique condition can be written as:

$$ \sum_{j=0}^{k} w_iX_j \space < \space \mu_{X_1, \dots, X_n} $$

Since the number of features appearing in the condition $k$ and the weights of those features $w_i$ and the features themselves and the right member of the inequality can all change, this means that there is a very large number of possible conditions to split on, each having an associated measure of impurity that has to be calculated. How do the implementations of oblique decision trees go about addressing this issue?

$\endgroup$
1
  • 5
    $\begingroup$ It strikes me that there may be more than one implementation of this, & they may address this issue in different ways (I don't know that, it's just possible). As a result, this question has a potential ambiguity that people might say requires clarification. Other than that, I don't see any reason this Q would be objectionable. $\endgroup$ Jun 28, 2023 at 0:18

1 Answer 1

7
$\begingroup$

For all practical purposes, as gung - Reinstate Monica noted, these sort of queries can be positively entertained by the community. I am not seeing any reason it could be closed as off-topic.

But that might not deter one to potentially vtc the question for it being too broad/or needing more focus. I would be more specific with my query along with proper citation of sources. After that I would be having no qualms with it.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .