PopularityPercentage meaning

[giuliowaitforitdavide]

As we already discussed, the idea is to use this as score to then segment the items/user based on popularity. Don't you think is more correct to put this class in a separate file? I don't think it's semantically correct to consider it as a segmentation technique

[mmosc]

Yes, you are right. We should come up with a name for the new file/class. I think popularity could be misleading, since InteractionSegmentation and ActivitySegmentation are also measuring "popularity", somehow. What about renaming the class InteractionPercentage and creating a file named percentages.py?

[giuliowaitforitdavide]

Maybe we could:

• split the test_pattern function in utils from the matrix function, and call the file matrix.py
• rename the test_pattern function in check_pattern to avoid misunderstanding with the tests function and put it in a errors_utils.py or inside the errors.py itself
• create another file (as you suggest) called segmentations_utils.py and put the PopularityPercentage function there. It should not be considered as a subclass of the Segmentation one, I think it could simply be a function.

Another idea could be to completely reformat the directories in this way:

recsyslearn
└─── errors
│   │   errors.py
│   │   utils.py
│   
└─── metrics
│   │   Entropy.py
│   │   ...other metrics
│   │   utils.py
│
└─── segmentations
│   │   ActivitySegmentation.py
│   │   ...other segmentations
│   │   utils.py

Do you like these options?

[mmosc]

I like the second option, but I think the last folder should not be called segmentations, but something like activity. PopularityPercentage and InteractionSegmentation are both measures of popularity/"activity" of the item, but the first one is not a segmentation.

[giuliowaitforitdavide]

The main idea is to put into segmentations all the possible segmentations for a given dataset, and move the functions such as PopularityPercentage into the utils.py because it can be considered as an helper for the segmentations. I don't know if it's clear and you like it

[mmosc]

I think PopularityPercentage should be considered as a measure of Popularity, as much as InteractionSegmentation is. They are at the same level in terms of usage of the library (they both assign a score of popularity to each item, one as a percentage, and one as a kind of "percentile"). Therefore I would not put PopularityPercentage in utils.py.