how we plot decision tree using fast tree Trainer?

[nighotatul]

@eerhardt - we got score and metrics but how we plot decision tree using trainer.

[eerhardt]

I'm not sure what you are asking for here. LbfgsLogisticRegression is not a tree based algorithm. It uses logistic regression to make classifications.

In general, if you want to get the underlying information about a model, you use the Model property on the transformer that is returned from fitting a trainer.

// Define the trainer.
var pipeline = mlContext.BinaryClassification.Trainers.LbfgsLogisticRegression();

// Train the model.
var transformer = pipeline.Fit(trainingData);
IReadOnlyList<float> weights = transformer.Model.SubModel.Weights;
float bias = transformer.Model.SubModel.Bias;

These weights and bias numbers are what is used to make the decision in LbfgsLogisticRegression.

[nighotatul]

@eerhardt - i have prepared sample on this below link.
this is Visualize sickit-learn decision trees with d3.js
http://bl.ocks.org/fractalytics/raw/495b63cf671b4c487bc40801366384e0/

we do not know python script but we know C#. so that please helps us to draw above tree using ml.net using score,weights,bias.
if any sample is their please share with us.

[eerhardt]

So the part that still confuses me is that you are explicitly asking about LbfgsLogisticRegression, which is not a tree-based algorithm. So there is no way to create a tree structure like your example shows.

The way logistic regression works is it takes the weights from above, and multiplies them by the values being entered as the input, plus some bias. If that resulting number is greater than some threshold (like 0.5), then it predicts true, if the result is less it predicts false.

Are you asking how to create a graphical representation of a tree using C#? Or are you specifically asking how to make a decision-tree out of a LbfgsLogisticRegression model?

If your question is the latter, I don't know. Like I said, LbfgsLogisticRegression isn't a tree-based algorithm. You would need to use a decision-tree algorithm (like LightGbm or FastTree) to get that information.

[nighotatul]

@eerhardt - Sorry for that confusion,then how we get visualization using fast tree algorithm?

[eerhardt]

To get the tree data from a decision-tree model you can use the .Model property on the Transformer. From there, you can grab .TrainedTreeEnsemble which has all the tree data.

How to actually use that data to visualize the tree is beyond my knowledge. Maybe @codemzs, @ganik, or @yaeldekel can answer that question.

[nighotatul]

@yaeldekel - i have prepared sample on this below link.
this is Visualize sickit-learn decision trees with d3.js
http://bl.ocks.org/fractalytics/raw/495b63cf671b4c487bc40801366384e0/

we do not know python script but we know C#. so that please helps us to draw above tree using ml.net using score,weights,bias.
if any sample is their please share with us.

[eerhardt]

Closing in favor of #4275.