On the 30th of October I presented at Data Sciences Speakers Club the talk “What is model complexity?”. This presentation was a quick introduction to the topic of model complexity.
Below you can find the topics that I covered during the talk:
- I started with the activity to the audience: What would you consider to build a model that estimates the probabilities of survival of a passenger from the Titanic? Would it be a regression model? A Decision Tree? A Neural Network?
- The ChatGPT definition of model complexity: Very interestingly the definition points out to two main components discussed: The structure of a model and the number of parameters. Also quite good to see that ChatGPT 4.0 now even includes a diagram of model complexity as number of parameters are included
- The main topic to highlight is: model complexity starts before building a model. As we are preparing and consolidating the data for modelling we are adding conditions and criteria to the model that will affect the design and outputs of the model.
- Measuring model complexity: Different techniques have different parameters to consider. The parameters in simpler techniques like regression or classification trees are easier to understand than in more complex techniques like gradient boosting or neural networks.
- One way to measure model complexity is if the model is easy to understand and explain. If not, it might be already too complex!
Below you can find the slides that I covered:
For more details about Data Science Speakers club and to attend future meetings find it in here
Cheers,
Eduardo
Speak the Data.com 