7.5 Model fit

After the previous steps features are available and can be fed into models.

Depending on the task certain models will be better suited than others. However, there is no clear rule when to make use of a certain model given a specific task. Therefore at the beginning it is not unusual to compare several models to find which algorithms are more promising than others.

Model fitting:

  • Fit several models
    • diverse models
    • speed up process
      • reduced complexity models
      • reduced data set
  • Identify best suited models
  • Increase complexity
  • Reiterate feature engineering if necessary
  • Favorite models identified
    • next step: tune model’s hyperparameter, see 7.6

The process is iterative as shown in the graph below

The next step is hyperparameter tuning of the model.