search

heitor57 / poi-rss

:european_castle::classical_building::convenience_store: Point-of-interest recommenders and diversifiers.
MIT License
1 stars 2 forks source link

Yelp Dataset 2021 version incompatibility #2

Open kayevalerie opened 2 years ago

kayevalerie commented 2 years ago

In the latest version of the Yelp Open Dataset, the metropolitan areas were changed, so the cities_checkin_data dictionary doesn't get populated when running datasetgen.py. May I request for access to the same version of the dataset that was used during development? Thanks!

heitor57 commented 2 years ago

I'd love to do this, but from what I've seen of Yelp's terms of use, I'm not allowed to distribute their data. I want to contact Yelp soon for more information on sharing the dataset I used. If I get any good news, I'll share it here.

heitor57 commented 2 years ago

@kayeval finally, I got permission to share the dataset I used, you can download it from Google Drive, and let me know if you have any problems.

kayevalerie commented 2 years ago

@heitor57 Apologies for the late reply, but thank you so much, I was able to access the dataset with no problems and was able to train a new model using it.

On another note, I was just wondering what steps I should take in order to add a new model as part of the existing recommenders (so I can run the same evaluation metrics on the new model)?

heitor57 commented 2 years ago

If you want to add a base recommender with the following steps below you should have no problems, if it's a post-processing model the process is extremely similar. I recommend that you consult the implemented recommenders to clarify other doubts.

Steps: (1) Create the configurations in the RecRunner class to manage the inputs; (2) In the RecRunner class, in the static method get_base_parameters, the recommendation model must expose its default parameters by a dictionary; (3) Set and create a handler to execute the recommendation model in RecRunner BASE_RECOMMENDERS attribute; (4) Set in CITIES_BEST_PARAMETERS dictionary at constants.py the parameters of the method to use in each dataset; (5) Set in RECS_PRETTY a name of the recommender to appear in visualizations.