Add optional exercise "Trial and Error" at the end of episode 2

[CunliangGeng]

Here is the exercise

Trial and error: play with your model (breakout, 20 mins)

Define a metric to measure your model's performance.

There are various strategies to try to improve the performance of your model:

• more data and better data
  • use more data
  • standardise the data
  • make the data more balanced
  • ...
• change the model architecture
  • add more hidden layers
  • add more neurons for hidden layers
  • use different types of hidden layers
  • ...
• change hyperparameters
  • use more epochs
  • ...

Choose one or more of the strategies and try it to see if it can improve your model's performance or not - how does this perform on your performance metric.

[svenvanderburg]

This issue is quite old, and I think at this to me episode 2 actually looks complete. In my vision, episode quite superficially introduces the full deep learning workflow. In episode 3 we really expand it and think about strategies to improve the performance of the model, there we actually have these kind of exercises.

Adding this exercise to episode 2 would only unnecessarily lengthen it, whereas the kind of skills that are tested with this exercise are already tested in episode 3. I'm therefore closing the issue.