TACKER:  koenig (Greg Koenig)
SUBJECT: .. To learn or to drink
DATE:    13-Mar-18 18:07:23
HOST:    faeroes

I currently work as a data scientist.  Much data science is done in Python.

If you're getting started, I'd skip TensorFlow initially and instead focus 
on scikit-learn.  One advantage of scikit-learn is that it has very good 
documentation that has a lot of examples.

A second advantage of scikit-learn is that the models are more or less 
interchangeable.  When you develop a model, you tend to have a few main 
sections in your code:

1.  A section that reads in your data and sets up data structures that 
comprise your feature vectors.  This is in my experience the largest 
section of code.

2.  A section that calls the machine learning model with your training 
data.  In scikit-learn, this is probably 10-ish lines of code.

3.  A section that evaluates the efficacy of your model with your test 
data.

Having interchangeable models means that you keep #1 identical and 
typically only have to change one line in #2 (the actual call to the 
model).  Professionally, the advantage of this is that you can quickly try
a lot of different models on a given problem to see if you get better 
results with different models.  As a student, the advantage would be the 
same although you'd specifically want to do this for the learning 
experience involved in order to get an intuitive understanding of what 
models work best for what kinds of machine learning tasks.

TensorFlow is a pretty sophisticated framework that requires a lot of 
intuition in doing feature engineering in order to get good results.  I 
personally would not start here.

In order to actually do machine learning, you are going to need a dataset 
big enough to support training and testing and then a related stream of 
data that you can do predictions on.  It's okay to use sample datasets for 
this (e.g., the famous Iris data) but I'd encourage you to try to find a 
use case that you care about personally because it will help to better 
engage you in the learning experience.

Good luck!  Data science is a very fun field!