Create Sample tool Usage

Question

I am using 'Create Sample' in my Model Training Workflow and not able to figure out right usage of 3 anchors from 'Create Sample'.

1. Estimation Anchor to 'Assisted Modeling'

2. Holdout anchor to 'Predict Tool'

3. Validation (Where to plug this anchor as validation set is for tuning model)

Thanks for assistance in advance

Create sample.png

alexnajm · Answer

@waqashussain yes I understand the concept of splitting the data to train versus test a model. I think to answer your question, there’s not a way to pass new data into the same model to change the model - you would just bring down another Predict tool and connect the same model output into it with the third dataset to further validate the model with new data. You can always go back into the Assisted Modeling tool to try out new parameters and build new models to compare side by side.

waqashussain · Answer

Hi Alex,

Thank you for your response.

As a fundamental principle of data science, a certain percentage of input data is allocated to the training/estimation set, while another percentage is allocated to the validation dataset for parameter tuning. Additionally, a holdout dataset is reserved to assess the performance of the model.

In the workflow I've shared, I've connected the estimation anchor to the model. However, I'm unsure where to integrate the validation anchor for fine-tuning the model. Could you provide guidance on this?

alexnajm · Answer

I am sorry, but I don't understand the question. Your workflow looks correct - typically people will have 80% come out of the E anchor for training purposes and 20% come out of the V anchor for testing purposes (which I think you have as your H anchor right now). Unless you need that extra dataset for further validation?