Improving the Usability of ML Operators

[SarahAsad23]

After teaching ML to a group of students using Texera, we noticed that it is not very intuitive to have to split the data twice when creating a ML workflow.

The spits produce:

1. Training data - passed to ML operator
2. Test data - passed to ML operator
3. Prediction data - used to do the predictions

For reference, here is the workflow that was used:

I am thinking it would be more intuitive for the train/test split to happen internally within the ML operator, which could remove the first split.

[chenlica]

There are two parts in this design that are not intuitive:

1. The ML (linear regression) operator has two input ports, one for training instances and one for testing instances.  A more intuitive design is to have a single port for the instances;
2. The output of the ML operator includes both the model and the "ground-truth instances."

I believe we need to have a better design.

[SarahAsad23]

After internal discussion, we decided that for future teaching scenarios we will use the "Training Linear Regression" operator instead of the original "Linear Regression" operator as it is more intuitive.

The Training Linear regression operator takes training data as an input and outputs a model. The Linear Regression operator takes train and test data, internally does the training and evaluation, prints the accuracy, and outputs a model.

We will later decide if we would like to possibly remove one of the operators.

[chenlica]

For now we will keep the "Training Linear regression" operator. We will decide later whether we want to remove the "Linear Regression" operator.