Part 1: ML Training Practice
One of the things that makes a very good Sommelier is their ability to figure out as much details about a wine as possible with very little information.
You are tasked with making a Sommelier program that is able to figure both the type and quality of wine from available chemical information. Also, you have a “flavor-ater” machine that makes a linear combination of multiple chemical features together (similar to PCA), which is counted as one chemical feature after combination.
A good Sommelier uses as little information as possible to deduce the quality and type. So, what is the best model(s) you can build for predicting quality and type of wine based on the least amount of features? What features should you choose?
Part 2: ML Project Walk-through
Create your own machine learning experiement! Begin with a problem in your field; go through the available/your own data, determine what type of problem it is, and discuss why machine learning could be a good solution for the problem. Research/quantify the baselines in the field for the task (remembering our discussion on ML validation methods), and determine a list of possible features of your data.
If we were to help collect data together, how can we best collect a representative sample? How expensive (resources, monetary, or temporal) would it be? What are some ethical issues?
Select the features in the data available to you that would be most relavent (this time you are not trying to minimize the features, but select the most appropriate ones), and the model/training mechanism you think would be most appropriate.
Finally, present your thinking! Share with us a few (1-3) slides on Friday afternoon. If you have additional time, possibly train the model on baseline data!