[ML][Inference] adding tree model #47044
Conversation
|
Pinging @elastic/ml-core |
|
@valeriy42 Let me know what you think. Additionally, I cannot find a use for BOTH |
| if (treeNode.getRightChild() != null) { | ||
| toVisit.add(treeNode.getRightChild()); | ||
| } | ||
| } |
There was a problem hiding this comment.
you could add an easy check for disconnected nodes by checking visited.size() == nodes.size()
The difference is |
| * @param decisionThreshold The decision threshold | ||
| * @return The created node | ||
| */ | ||
| public TreeNode.Builder addJunction(int nodeIndex, int featureIndex, boolean isDefaultLeft, double decisionThreshold) { |
There was a problem hiding this comment.
So the nodeIndex variable here is what is called split_index in JSON. We should homogenize the names.
There was a problem hiding this comment.
I think node_index is a more friendly name. Honestly, the value will probably be passed down from the tree as the node array order needs to be guaranteed for serialization to work well.
| new NamedWriteableRegistry.Entry(PreProcessor.class, FrequencyEncoding.NAME.getPreferredName(), FrequencyEncoding::new), | ||
| new NamedWriteableRegistry.Entry(PreProcessor.class, OneHotEncoding.NAME.getPreferredName(), OneHotEncoding::new), | ||
| new NamedWriteableRegistry.Entry(PreProcessor.class, TargetMeanEncoding.NAME.getPreferredName(), TargetMeanEncoding::new), | ||
| // ML - Inference models |
There was a problem hiding this comment.
I tried to avoid using "models" since there is a long tradition of overloading this term. In JSON schema the section is called evaluation. I am not particularly invested in the term, but we should stick to the same terminology everywhere.
This also may mean that we rename "evaluation" to "model" in the JSON schema.
There was a problem hiding this comment.
I really like using model as it is a unified name with the ensemble and its nested models object.
It seems logical to me that a model that is an ensemble will have many models. If we choose to use evaluation I think ensemble should have an evaluations field.
* [ML][Inference] adding tree model * renaming features for updated schema
This adds the base tree model that can be used for our future ensemble model.