This repository contains our code, used in the TMDB Box Office Prediction competition.
The data was obtained from TMDB.

[Optional] Create a virtual environment.

$ virtualenv --python=python3 --system-site-packages boxofficevenv
$ source boxofficevenv/bin/activate

Then, install the required dependencies.

$ pip3 install -r requirements.txt

Note: you might need to replace pip3 with pip.
Note 2: this does not mean the code is Python2.x-compatible (if it is, it's completely coincidental).

train.csv and test.csv contain the training and test data.
fixes_train_budget_revenue.json contains fixed budgets and revenues for some movies (keys represent imdb_id of movies) in the training set.
fixes_genre.json contains genre information for the examples in training and test set which do not have it.

Using root mean squared logarithmic error (RMSLE) as it is used for submission scoring on Kaggle. The approaches are briefly described in the next section.

Baseline: predict mean revenue of movies in training set for the movies of test set. 70%-30% train-test split used for offline evaluation.

Basic features: use 3 basic features: genres, original_language and runtime. Encode genres and original_language using 1-hot-encoding. Add new feature (count of genres provided for film). Group less frequent original_languages (< 5 occurences in training set) under 'other' attribute. 70%-10%-20% train-val-test-split used for offline evaluation. Best score obtained with XGBoost.

All features: use all of the features: runtime, popularity, budget (excluded budgets < 100), original_language, spoken_langugate, important_cast_count, top_director/writer/producer, production_companies/countries_count, is_weekend, release_weekday

All features: use four original features: popularity, runtime, budget (excluded budgets < 100), and genres.