What is your analysis plan? How will you present your results?

• Good test coverage in your scripts! Although there are a few functions without tests.
• Your final report draft is good, the question, analysis and results are clear, but elaborate more on future directions: which models could be more robust?

• makefile graph - added by Simar
• create docker for whole project
• address feedback

• Below are some thoughts about the visualizations in this report and some suggestions for improvement, and notes when things are done well. You should implement these in milestone4
• • Great job adding figure numbers and captions! Though your captions could have been a bit more informative, overall this is a very good practice. Also, both figures are called Figure 1
• • Figure 1
    • • This is a really effective plot, great job!
    • • I would make the plot a bit smaller, and the textsize a bit bigger so it's proportionate to the rest of your report
• • Figure 2
    • • consider combining Figure 1 and 2 together in a side-by-side way and call them A, B so it's easier to interpret, read, and takes up less space
• • I think your figures are very reasonable for your project and research question. Nice work!

A more thorough explanation of the data and the ELO ranking system is required to be able to understand your project goal.

URL:
https://hub.docker.com/repository/docker/tbarasch/g306_522
to be added as collaborator reach out with docker ID

We find that both models achieve similar results, with 0.771% accuracy

for the logistic regression and 0.767% accuracy for the random forest

classifier. Which overall is a pretty unreliable prediction model, casting doubt over the method presented by FiveThirtyEight.

0.8% ? That is pretty low... are you sure there isn't a typo in your report ?

You can improve your repo structure (you have CSV files and figures in your scripts folder, for example)

After re-running your EDA script, you in fact load a marmap library that isn't used.

Explain model performance
Explain ELO rating
Add some graphics

Continues thing - try to make sure you write comments into your code as you go, its pretty hard to read someone else code if you don't know exactly what was going on, and comments can help a lot.

Your model.py script didn't run (problems saving figure in line 77)

Since the objective is to find out whether ELO can be a good predictor for game results, and we have 3 classes, it's clearer for the readers if we show precision and maybe recall.

You should download and save the data using a script according to Milestone 1 instructions.

What is the motivation for selecting the specific models you have listed in your proposal?

Great job overall! Fix all remaining issues for milestone 4, and ensure your analysis runs on all computers before submitting.

• Missing test and author info on eda.py

Some details about your report:

• Your equations aren't rendering properly.
• As discussed in the lab, elaborate on discussion and future directions (why is ELO rating not a good predictor?)

• Since the data is based on football and elo ratings related to it , so it will be good idea to add some information about how elo rating is related to the game.

Always render a .pdf of .md version of your report so it can be read in Github.

Great work overall!

Mechanics:
Your dependency graph is not in your readme. Your use of issues was very basic.

Quality:
Missing description on top of docker file with authors and explaining what it does.

Vis:
Most feedback was not addressed.

Writing:
Equations still did not render properly.

• Proofreading is required to correct grammar mistakes and typos.
• Pay attention to communicating through Github issues and writing more distinct commit messages.

Use GitHub issues to communicate and close after addressing them.

Split your data into training and test sets.

Need to add the docopt compatibility to all scripts

• No big issues that need to be fixed for Milestone 4, good job!
• Remember to use issues for communication on this last milestone. and close the existing issues after addressing them.
• Elaborate on your conclusion/discussion, your future directions are still vague. Why could the suggested methods work better?
• Using wrong release link in personal submission.