Peer feedback

[trevorki]

Hi! Your project looks good and you got some interesting results. Here are some things I noticed while reading that you might want to think about:

README file:

• The conda environment did not install for me… this could be my issue as conda environments have a 50% success rate on my computer. Hopefully a Docker container will fix this.
• There are no instructions for running the Makefile.

Report: Summary:

• You say “We were more focused on reducing false negatives to capture more revenue.” You should clarify this connection. How are false negatives related to revenue?

• If the F1 score is not good enough, what score would you consider to be good enough?

• You mention that “we were using the f1-score as the main metric but had a bias of recall over precision”, What does this mean? That you will use F1? Recall? Use recall as a tie-breaker in the case of equal F1 scores?

• The reference the article on telemarketing is from 1991, which seems out of data considering how our relationships to phones have evolved since then.

Results & Discussion:

• Since you refer to features like previous outcome, Euribor 3 Month Rate, Employment Variation Rate, etc, it would be nice to see a description of what these mean.

• In Figure 3 the plot legends say Not Purchased/Purchased but the figure caption refers to Subscribers/Non-subscribers. Consistent terminology would be more clear.

• Figure 4 and figure 5 don’t render correctly in the .md report when seen on github.

• When referring to sklearn parameters like max_iter, the parameter names should be in a different typeface (italic, etc) for readability.

Good luck with the rest of the project!

[ChuckHo777]

Thanks for the feedback, will look into it.

[anodaini]

Addressed! Thank you Trevor!