Peer Review

[asneider]

Cleaning

• Why did you not use red or white as a boolean feature when you combined the datasets? You really have one dataset with a feature breaking it up into two.
• I’d suggest a little more data exploration. Just because no data is missing doesn’t mean you don’t have bad data or outliers. It seems strange no red wine is rated above an 8. Do you have a lot of really good or really bad wines in the dataset?
• I don’t see standardization used in the code although the thought seems to be there.
• You mentioned correlated features when we spoke, but I don’t see any code handling them
• I’d be interested to know if how red vs white affects decision to use one correlated feature vs another

Visualization

• Standardizing features before boxplot will improve visualization

Models

• I don’t see any of the modelling/analysis so it’s hard to give further feedback
• Using random forest on entire dataset could help determine if red/white feature is relevant to quality equation
• Is PCA useful when there are just 11-12 features?

Other Comments

• I liked your research on what the features mean. I think you can utilize it in your feature selection and possibly feature creation.
• Additional comments/observations throughout the notebook would be helpful for though process

@masongallo @lemonsoup