Got it — sounds like you’re trying to bypass the whole “training” aspect and just hard-code your decision logic in a tree-like form. In that case, sklearn’s DecisionTreeClassifier isn’t really the right tool, since it’s built to learn from data. A custom tree structure, like the Node class example given, would give you more control and let you directly define each condition without needing any training step. This way, you still get the decision-tree behavior, but exactly how you’ve designed it.