Startup task: issue#2 attempt#1

[urvigodha]

#2 Startup task: Train and Test on winequality dataset using random forest classifier
Hi @dzeber and @mlopatka, this is my first attempt on the startup task.
Would really appreciate your feedback. Thanks!

[mlopatka]

Thank you for the contribution @urvigodha.

This is a great start on the classification task. I'm going to merge it in as we are close to the application deadline. It is good that you observed the perfect separation possible due to the quality/recommend relationship, performance excluding the quality feature is more realistic for this task.

1- I liked your inclusion of a transformation function to normalize the features to a comparable magnitude. It would have been helpful to have a bit more context/interpretation of what this means for performance on future data. How would you implement a comparable normalization for a new wine that is introduced later on in a production deployment.
2- Further discussion on your choice for a 0.3/0.7 testing/training split would be helpful. Do you expect that random sampling in this case is skewing the performance results?
3- The random forest algorithm has many parameters that can be tuned, further exploration of these is interesting in quantifying the stability of the performance reported here.

thanks again for your contribution.