Warning
This page is located in archive.

Scoring

The evaluation is divided as follows:

  • Automatic evaluation tests the correctness of the found strategy (matching your returned actions with optimal actions for all states) in several different environments (possibly with different discount factors).
  • Manual evaluation is based on code evaluation (clean code).
Evaluated performance min max note
Quality of value iteration algorithm 0 2.5 Evaluation of the algorithm by an automatic evaluation system.
Quality of policy iteration algorithm 0 2.5 Evaluation of the algorithm by an automatic evaluation system.
Code quality 0 1 Comments, structure, elegance, code cleanliness, appropriate variable naming…

Automatic evaluation:

  • policy match 95% and more (average on n tested mazes): 2.5 points
  • policy match 90%-95% : 2 points
  • policy match 85%-90% : 1.5 points
  • policy match 80%-85% : 1 point
  • policy match 70%-80% : 0.5 points
  • less than 70% match: 0 points

Code quality (1 point):

  • appropriate comments, or the code is understandable enough that it does not need comments
  • reasonably long (or rather short) methods/functions
  • variable names (substantive names) and functions (verbs) help readability and understandability
  • pieces of code do not repeat (no copy-paste)
  • reasonable memory and processor time saving
  • consistent names and code layout throughout the file (separate words in all methods in the same way, etc.)
  • clear code structure (avoid, for example, unpythonic assignment of many variables in one line)

You can follow the PEP8 (Python style guide). Most editors (including PyCharm and VS Code) warn about PEP8 deficiencies themselves. You can also get inspired e.g. here or read about idiomatic python on medium or at elsewhere.

courses/be5b33kui/semtasks/03_mdp/scoring.txt · Last modified: 2024/03/26 20:33 by xposik