ahlag / workout_recommender Goto Github PK

ASNAPP is a workout video recommender web application. This project was completed as part of the UC San Diego DSC180B Capstone.

Authors (sorted alphabetically by first name): Amanda Shu, Najeem Kanishka, Peter Peng

The website is no longer live. Please see our video demonstration: https://www.youtube.com/watch?v=QJFg0HguGuI

The data is scraped from https://www.fitnessblender.com/. We are using the data for academic purposes only.

• run.py: Run to get data and model results.
• app.py: Runs flask web application.
• workout_db.sql: Contains sql statements for creation of tables in database.
• requirements.txt: Python packages required to run project
• wsgi.py & Procfile & runtime.txt: Entrypoint for Heroku, used in website deployment

Source

• The src/data folder contains scrape.py, the web-scraping script that writes three data files into data/raw folder. fbpreprocessing.py takes these raw data files and outputs cleaned/transformed data files into data/preprocessed folder. youtube.py grabs youtube related data from the Youtube API. model_preprocessing reads in preprocessed data and transforms the data into what is needed for model inputs.
• src/models contains run_models.py which trains and evaluates the models. Models are implemented in lightm_fm.py and top_popular.py
• The src/utils folder has clean.py which implements the standard target clean.
• The src/app folder holds files for the web application. forms.py contains wtforms classes for registration/login pages. recommendations.py holds code for filtering user preferences and building recommendation lists. register.py contains helper functions to create the sql insertion/update statements for registering users and updating their workout preferences.

Static:

• Theimages folder holds a gif (source) used for the loading page. No copyright intended.
• The js folder contains several javascript files. overlay.js is for the display of the popup videos on the recommendation page. workout_info.js is for registration and update preferences pages. rec.js is for the loading page and recommendation engine logic on the recommendations page.
• libraries/slick has several files for the carousel, styles has a css file, and favicon.io is the dumbbell icon
• vendor holds several javascript files (Bootstrap, JQuery) for styling/theming of the website

Config: data-params.json has file paths outputs for data collection/preprocessing and test-params.json has the data paths for the test target. To webscrape, this folder should also include chromedriver.json. To gather Youtube data, api_key.json specifies the api key. To run the app, db_config.json has the database configurations.

Notebook: eda.ipynb is a notebook with exploratory data analysis on scraped data. param_comparision.ipynb is a notebook reporting the recommendation models' performance across a couple parameters. top_popular_extension.ipynb is a notebook looking into adding Youtube API data into the top popular recommender. KNN_collab.ipynb contains results of a KNN collaborative filtering model from surprise package and a pure collaborative filtering from LightFM.

Templates: Holds html files for the various endpoints.

Testdata/raw: These are fake datasets meant to be used with the test target.

Docker: Docker related files. See here

Materials: Contains pdfs for presentation slides and a report detailing our methods/implementations.

There are two ways to run this project: a) Docker (preferred) or b) Locally
a) To Run in Docker:

1. Pull the container with docker pull nkanishka/workout-recommender
2. Run the container using:

• General Use (most likely): docker run -it -p 5000:5000 workout-recommender
• UCSD DSMLP Servers Only (only for DSC180B students and faculty): launch.sh -i nkanishka/workout_recommender_dsmlp -c 4 -m 8

kubectl port-forward <Kubernates Cluster Name> 5000

ssh -N -L 5000:127.0.0.1:5000 <AD Name>@dsmlp-login.ucsd.edu

3. Inside container/cluster, type cd Workout_Recommender. Note that in the DSMLP environment, you will need to manually clone this repo.
4. If using website, go to localhost:5000
   

b) To run locally, install requiremnents.txt into a virtualenv. Make sure you have Python 3.8+ and Pip installed.

• To get the data, run python run.py data. This scrapes the data and cleans the data and saves these files into /data/raw and data/preprocessed respectively.
  • Note: for scraping, this assumes that there is a file config/chromedriver.json that specifies where the path to the downloaded chromedriver.exe file for your Chrome version lies in the attribute chromedriver_path.
  • Note: for making requests to Youtube API, this assumes that there is a file config/api_key.json that specifies the api key in the api_key attribute.
• To run model results, run python run.py model. This takes in the preprocessed data, trains the models, and prints out the NDCG scores for each model.
• Standard target clean is implemented, and it will delete the data folder.
• Standard target all is implemented, and it equivalent to running python run.py data model.
• Standard target test is implemented, and runs the data preprocessing and modeling results on the test data. The purpose of this target is purely to check the implementation of the code.
• Use python app.py to run the app locally.
  • Note: this assumes that there is a file config/db_config.json, which has database host, user, password, and name information.
  • And a file config/flask_keys.json which has a Flask secret key