Which predictor gives best accuracy?

I want to compare the accuracy of several learners/predictors and find out which gives the best accuracy. I have loaded the data, feed it to a Partioner, and obtain the training and test sets. After that I add a learner (e.g Random Forest), its predictor, and finally a Scorer. And I get its accuracy, fine and good.

However, I want several different learners/predictors, and can easily add them, each with their own Scorer giving its accuracy.

Then the problem: I want to feed all Accuracy scores to a node, which tells me which one was the best. I cannot find a node that accepts multiple inputs and combines them into a table that I can sort and find out the maximum. There is Joiner, which accepts two… but multiple?

Hi @ssq,

Appender Node with multiple entries may be a solution:

Hope this helps.

Best

Ael

1 Like

Hi @aworker,

Indeed it would work, with some column filtering and possibly renaming to get sensible output. Perhaps Concatenate could also be used, I found?

Ideally, the Predictor should feed its name forward, and at the end one should get a table:

Row ID Accuracy
Random Forest. 0.95
Gradient Boost 0.93
Decision Tree 0.90
Simple Regression 0.85

That would be perfection!

Is it possible to set a flow variable? to a string (the name) and have it propagated?

As you notice, I’m new to KNIME… :slight_smile:

Hey @ssq,

If you’re building a classification model for this use case, you may want to try our AutoML verified component and see if it fits your needs. There are a number of models it can train and there are several related components on the hub for eXplainable Artificial Intelligence (XAI) and other purposes.

Cheers,

@sjporter

3 Likes

Yes, I will do so. Thanks @sjporter for the tip!

1 Like

When you want to append the name then use a constant value column in each ml model and write the name. Other option would be to create a variable with the name
br

1 Like

This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.