Implication of the Results Obtained for Football Problem

Course Q&A Deep Learning Specialization Improving Deep Neural Networks: Hyperparameter tun
1

Hello,

After completing W1A2, I had a doubt. We get to the point where we achieve 95% accuracy in the test set, along with a graph which shows the proper classification of the red and blue dots.

image
image877Ɨ463 4.21 KB

(sorry for the MS paint version; at the time of writing this post I had closed my notebook so I would have to run the notebook upto that point over again :sweat_smile:)
So what I am unable to understand is how will this graph help the football team in real time?

Thanks,
Sushant.

1 Like
2

Please share the following details:

  1. Course / week this question belongs to.
  2. Link to the assignment.
  3. Definition of football problem and what the graph means.
1 Like
3

Hello,

  1. Course 2 Week 1 Assignment 2
  2. Coursera | Online Courses & Credentials From Top Educators. Join for Free | Coursera
  3. Welcome to the second (required) assignment of Course 2 of the Deep Learning Specialization! In this notebook, youā€™ll apply L2 regularization and dropout to a deep learning model that recommends positions to football players. Your ā€œgoalā€ (pun intended) is to avoid overfitting and make better recommendations to the goal keeper.
1 Like
4

It is the Regularization assignment (A2) from DLS C2 W1.

The decision boundary analysis will not help anyone in realtime. This is what we do when we are training the model to understand when we have gotten our hyperparameter tuning to the ā€œGoldilocksā€ level: not overfitting or underfitting but ā€œjust rightā€.

Itā€™s always the case with supervised learning that the training happens beforehand with the goal of producing a model that will work as well as possible when deployed in a real application with new data as input.

But this a bit more artificial case in that it would be sort of ā€œstaticā€ guidance to the goal keeper in terms of where to kick the ball in a real game scenario such that he has the best chance of a good outcome based on all the input training data about past occurrences of the situation. Mind you, I know nothing about soccer and would expect that in any sport related scenario like that the situation is incredibly specific to what is actually happening on the field and the goalieā€™s observation of the capabilities on that particular day of his own team and the opponents. So, as mentioned, this is kind of an artificial example. At least thatā€™s my take, which is probably worth exactly what you paid for it. :laughing:

2 Likes
5

But suppose in case that graph was real-time data. And somehow the goalie had real-time access to the graph! (Perhaps wearing Google lens or something like that :laughing:! (Iā€™m serious because I just heard that there are glasses equipped with camera which can capture images/video and send it to GPT-4 and give back data to the wearer, though in audio format)). So in that case how would the graph inform his decision?

1 Like