Project 5: Traffic or Exploratory Project

Milestones

Sample Solution

github.com/vault50/e80-ai-solutions/tree/master/5/traffic/solution

Please notify Doug if you don’t have access to the above via vault50.

Grading Information (Traffic)

Projects in CSCI E-80 are autograded along the axes of Correctness and Style. Teaching fellows grade as well along the axis of Design. Total scores on Projects are out of 15 points, weighted as follows:

• Correctness scores range from 0.00 to 7.50 (all real-numbered values), awarded by the autograder.
• Design scores are any of the values 0, 1, 2, 3, 4, 5 (integers only), awarded at the discretion of their teaching fellow based on some of the guidelines specified below
• Style scores range from 0.00 to 2.50 (all real-numbered values), awarded by the autograder.

Grading Information (Exploratory)

Unlike most projects in CSCI E-80, teaching fellows fully grade all three axes of this problem, which is somewhat like a final project inasmuch as students themselves design the specification for their project in consultation with their TF. Total scores on this Project are still out of 15 points, weighted as follows:

• Correctness scores are any of the values 0.0, 1.5, 3.0, 4.5, 6.0, 7.5, awarded at the discretion of their teaching fellow based on some of the guidelines specified below.
• Design scores are any of the values 0.0, 1.0, 2.0, 3.0, 4.0, 5.0, awarded at the discretion of their teaching fellow based on some of the guidelines specified below.
• Style scores are any of the values 0.0, 0.5, 1.0, 1.5, 2.0, 2.5, awarded at the discretion of their teaching fellow based on some of the guidelines specified below.

Per the specification for this project, students are expected to have done all of the following

• Solicit approval. Send a brief proposal email describing what topic they’d like to explore and what data set they plan to use.

• Use TensorFlow. Build a neural network using TensorFlow to perform a classification task based on the data set.
• Supply a README. There should be a comprehensive README file that describes, in one or more paragraphs, what the program does, and how to run it.
• Discuss their process. In the README, document the experimentation process.

This project is graded along the three of axes of Correctness, Design, and Style. Each should be treated as independently as possible (code that is very broken, but has some good ideas that didn’t come to fruition and is extremely well-styled should maybe earn a 2 for Correctness, a 3 for Design, and a 5 for Style, for example.)

Correctness (Exploratory)

Award:

• 5 if the requirements above are met with the student having introduced no bugs in their neural network.
• 4 if there is one minor bug.
• 3 if there are multiple minor bugs, or one major bug, or at least one of the requirements appears to be missing entirely.
• 2 if there are several major bugs, or at least two of the requirements appear to be missing entirely.
• 1 for a very clearly low-effort (missing the implementation of three or more requirements) or very broken submission.
• 0 for no submission at all.

Autograder Overrides (Traffic)

As a grader, you do have the authority to override the autograder in situations where things seem to have gone wrong (such as the student poorly organizing their code in a way that the autograder can’t find the file, or they have imported a library that is not installed by default in the autograder, etc.)

To do this, when grading for Design, which is normally where the grader’s work is confined, type the , (comma) key to jump back to the erstwhile-invisible 1: Autograder question, and modify the “Score Adjustment” you see there as appropriate. You can download the student’s code and test it locally at this point.

In general, the style score should not be adjusted here, and for each “point” of Correctness you would want to add (if it were scored as Design is), you add 1.50 to the figure there.

Design (Traffic and Exploratory)

When grading for Design, try to treat this as independently as possible from the other two axes. Code that is very broken, but has some good ideas that didn’t come to fruition and is extremely well-styled could maybe earn a combined 5.68 from the autograder and perhaps an additional 3 for Design, for example.

Design is primarily concerned with how elegant, efficient, readable, and clear the student’s code is. Some design questions to consider and leave feedback on include (but are not limited to):

• Is the code unnecessarily complex?
• Is code reused where possible?
• Are there unnecessary loops, conditions, and/or variables?
• Are there places where the student’s code could have been made more efficient?
• Is the code logical in structure?
• Is there repetition of code that could be more cleanly written with the addition of a function?

Award:

• 5 for the best designs, with absolutely no room for improvement.
• 4 for better than average designs, with maybe a few quirks.
• 3 for good, solid designs, though there are a few suggestions for areas for improvement.
• 2 for fair designs that feature significant room for improvement.
• 1 for poor designs reflecting very little effort or consideration of design implications of coding decisions.

A missing README.md file, required for both Traffic and Exploratory, incurs a 1 point Design deduction after you have established the score otherwise, above.

Style (Exploratory)

Style is concerned with the aesthetics of the code (whitespace, comments, etc.). Some style questions to consider and leave feedback on include (but are not limited to):

• Is the student consistent about the use of single quotation marks vs. double quotation marks?
• Is the use of spacing and indentation consistent?
• Are there appropriate comments in the code?
• Is the use of variable names clear, meaningful, and consistent?
• Do files have appropriate names?
• Are there places where an extra newline would help with readability? Are there places where there are too many unnecessary newlines?
• Do lines run on for a long time where it would have been clearer to separate into multiple lines?

Award:

• 5 for the best style, with virtually no room for improvement.
• 4 for better than average style, with maybe a few quirks.
• 3 for good, solid style, though there are a few suggestions for areas for improvement.
• 2 for fair style that feature significant room for improvement.
• 1 for poor style reflecting very little effort or consideration of aesthetics, commenting, etc.
• 0 for no submission at all.