Skip to content

Latest commit

 

History

History
49 lines (36 loc) · 3.03 KB

README.md

File metadata and controls

49 lines (36 loc) · 3.03 KB

Gross-Movie-Revenues

1. Data Sources

  1. Link to pull top movie titles per year along with their gross revenue: https://www.the-numbers.com/market/2022/top-grossing-movies
  • Web scraped each year's top # movies to create a dataset of movies
  1. Link to the API used to gather movie information based on the above movie titles: https://rapidapi.com/standingapi-standingapi-default/api/moviesdb5
  • Ran API calls to pull in the information for the movies found via the top charts. Not all movies were able to be found due to generic titles

2. Access Data

  1. Data is stored in PostgreSQL database. Instance of database is hosted on ElephantSQL Request accest to URL to view tables and information inside them. The current database size is 13MB while the max database size is 20 MB

    Tables to use:

    • moviesinfo: contains information on each movie from the MoviesDB API
    • movies2023: contains movies that are coming out in 2023 - these movies are used for predicitve purposes from the client
    • moviesgross: contains the movie titles and their annual gross box office revenue
      • look out for duplicate titles as some films are released into box office multiple years

    Additional tables created that can be deleted to free up space:

    • movieinfo: has unformated data for movie information, use moviesinfo
    • test: used to structure column sizes for moviesgross

3. List of Experiments Ran

Two Pipeline Model Setup files can be found under the model folder. "...Final to Serialize.ipynb" was the notebook used to create the serialized model.

  1. Ran LinearRegression, Ridge, Lasso, RandomForestRegressor default models on the data.
  2. Based on default models' performances, we moved onto selecting RandomForestRegressor as best model and did hyperparameter tuning via GridSearchCV.
  3. Model was serialized as pkl file but could not be pushed to GitHub as it exceeded 100 MB.

4. Todo items

  1. Load server to EC2 instance.
  2. Serialize (pickle) the models and load them to S3 bucket.
    • Create a SQL table to store the models and their scores that the S3 bucket would connect to
    • Every time new data comes in (e.g. every 2 weeks), score the different models on the test data and choose the model with the best R^2 as the model that connects to the front end (on Flask) to make predictions
  3. Use Cron to schedule web scraping, API requests, and adding information to database every 2 weeks
    • check for new movies on the-numbers.com that have been out for a specified duration (minimum 2 weeks) and add to database
  4. If time allows, find another API or database to pull better data

5. Configurations

  • Need to create a config.py file in the web_scraper/Files and src/server/model_prediction directories with a variable labeled api_key, storing your MoviesDB api key
    • this is an ignored file -> do not rename or else you will be sharing your API information publically on GitHub
  • Store your finalized pickled model in the src/server/model directory as the file name "model.pkl" for model predictions within the server