W1 - Regularization model

Hello everyone,

I have completed the notebook regarding the French Football team exercise. I just have one practical doubt; let’s say the manager is speaking to me and he doesn’t care how the model is developed, he just wants me to tell him where should the GK be kicking the ball, according to the model. How could I do this specifically? Asking differently, how do we go from the model being finalized to actually using it to provide an “English” output to the coach (“according to the model, on this specific game, you should ask the GK to kick it to the left side of the field; on that specific game, he should kick it to the central pitch”)?



Just didn’t want this to be forgotten down the list :slight_smile:

If anyone could help me out understand this, thanks.

Hi @jpedroanascimento,

Pretty fun question, actually.
Let’s think about it! Suppose we start with a trained model.

This model was trained with the position of the player that hit the ball, and the label is the team the player that hit the ball belong to.

So, the model tries to predict what team will head the ball if the ball is kicked to a given location (so the keeper knows where to send it). The input shape (you can just add print(train_X.shape) to the second cell to inspect) is (2, n) where the 2 is (x, y). The labels are (1, n) and it just 0 or 1 for each of the teams.

Say your convention is 0 = France and 1 = The other team.

The model expects a (x, y) input to return either 0 or 1. You, as the coach’s assistant, will input a vector or N positions (N pairs of (x, y)s) and from those, you can give back to the coach all of those positions that returned 0 (which is France).

Obviously, this example is super simple and doesn’t capture all of the nuances of the situation but hopefully illustrates the point. In real life scenarios, you probably have some data feeding the model in real time and displaying the probabilities of ball being received by a given player. Or even more realistically you’ll have a way more complex model that takes into account the current position of players themselves when the ball was headed (eg. not only who headed the ball but what other players where around) etc, etc.

There are endless possibilities into what you can do with ML in this scenario and they will only be limited by the data (and the quality of the data) you train the models with.

Hope that helps!