GitHub repo: shuchenku/PA2_movies2 https://github.com/shuchenku/PA2_movies2.git

CODECLIMATE REPORTS:

MovieData class: https://codeclimate.com/repos/54cac94ae30ba00498001947/MovieData MovieTest class: It was rated A but now CodeClimate fails to recognized that file.

My interpretation of the Codeclimate Grading:

Playing with CodeClimate gives me the sense that it emphisizes more on class simplictity rather than algorithm complextity.

My similarity function, which adopts a modified Cosine Similarity as metric, requires extra data structures and more lines of mathematical operations to be O(n). CodeClimate to did not like that and marked it as a high complexity function; the same is with my load_data() function where extra data structures were created to enable faster access to movies/users/rating. The way I designed the MovieData class is mainly focused on efficiency; if it was designed differently I imagine many of the methods will be O(n squared) complexity but CodeClimate might give me a better score. After considering the tradeoffs I decided to stick with what I have and be happy with a D (also ~ 1 min runtime).

MEAN ERROR Pearson Cosine 0.5 cutoff 0.5 cutoff u1: 0.83735 0.7959 u2: 0.82475 0.79305 u3: 0.8153 0.7912 u4: 0.80705 0.7916 u5: 0.81275 0.7966

RUNTIME Test size(u1) Runtime 10 1.0s 100 1.0s 1,000 3.1s 10,000 27.4s 20,000 54.8s 20,000(u3,4,5) ~70s

INDEX DEFINATION:

popularity index = 1/log(# of reviews)

similarity index = min(intersect(# of user1 reviewed movies, # of user2 reviewed movies), 8)/8 * cosine similarity Cutoff similarity index value for being "most similar": 0.5

* min # of movies required to not get penalized is set to be 8 b/c it gives the best prediction results.
* Cosine similarity improved prediction accuracy compared to Pearson correlation, therefore I cheaged the algorithm in Movies-2

Prediction of user u on movies m is based on average rating of u's similar users's ratings on m.