Get Posters

[kjprice]

• [x] Save a list of all the movies ids movies.id.to_csv(PATH)

Create a script that

• [x] For each movie id (example "19995") hit the API to fetch data about the movie: https://api.themoviedb.org/3/movie/19995/api_key={API}
• [x] For each JSON response, take the poster_path field
• [x] With the poster_path, download the image by using the url http://image.tmdb.org/t/p/w400/POSTER_PATH (example)
• [x] Store the image locally with the movie id (example 19995.jpg) in the filename
• [x] Move samuel feature code to its own python file (then run using %load filename.py)
• [ ] Find baseline for test/train/validation
• [x] Create train/test directories for directory - stratify on samuel feature. http://scikit-learn.org/stable/modules/generated/sklearn.model_selection.StratifiedKFold.html

  skf = StratifiedKFold(n_splits=2)
  skf.get_n_splits(movies.id, movies.samuel)

• [ ] Store all the images in S3 or Google Drive
• [ ] Create a new jupyter notebook that can go through all 5000 posters and will create a CNN model that can predict if Samuel L Jackson is in the movie

Optionally

• [ ] Before training the model with the posters, get a ton of images of Samuel L Jackson otherwise (and other images not of samuel l lackson) to help with the horrendously imbalanced dataset

[kjprice]

Build Model to predict Samuel L Jackson from images

• [x] Make sure that all images have the same dimensions
• [x] Load in movies dataset - create samuel boolean feature
• [ ] Create new dataframe that includes samuel and the entire image for each record
• [ ] Make sure that each poster where samuel == True includes samuel's face
• [ ] Create model that predicts samuel based on new dataset - expect results to be terrible

Possible ways to improve model

• [ ] Balance dataset by editing the samuel images in a bunch of ways (to create new images)
• [ ] Balance dataset by importing a bunch of pictures of samuel l jackson