Daphne Cornelisse
Mar 7, 2018 · 2 min read

Hi Mikko,

First of all, thank you for taking the time to write this comment, I really appreciate the effort you put in!

  1. The problem you desribed is not really a supervised learning problem.

The exact assignment was to build a wine classifier which recognizes the wine label based on 13 features. From what I have learned so far is that we have 3 main types of Machine learning: supervised, unsupervised and reinforcement learning. Given that we know the labels of the wine and we have 13 attributes, doesn’t that make this a supervised classification problem? I will talk about training in point 3.

https://medium.com/deep-math-machine-learning-ai/different-types-of-machine-learning-and-their-types-34760b9128a2

2. The input layer should have 13 layers instead of 178.

This is correct, thank you for pointing this out. I must have confused it when I was watching 3Blue1Brown’s video on Neural Networks. His input data consisted of 784 nodes which is 28x28 pixels.

3. Overfitting

I am aware of the problem of overfitting and the need to have a train, dev and test set. At first I considered mentioning the problem of overfitting and the fact that the accuracy is not reliable because not using a test set in the last paragraph. But in the end I decided not to mention it because I was afraid it could be overwhelming for people reading about Neural Nets for the first time and seeing all these new concepts.

Regarding the actual problem and overfitting; because this was the first week of the bootcamp, for the sake of keeping things simple, we did not have a training, dev and test set yet. These were introduced in week 2 and implemented this in our models from then on.

Thank you for wishing me luck! It is a lot of fun & learning a lot. :)

Best, Daphne

    Daphne Cornelisse

    Written by

    Neuroscience Student at Erasmus University College | Instructor Bletchley ML Bootcamp