• 

  Highest Accuracy of 94.3% w/ 0.056 error

Team Members

Paris Phan , Abhinav Pappu

Methodology

For our model, we used a four-layer Deep Neural Network. We used batch normalization and dropout to regularize our models. We derived new features like the Decade the movie was taken and log No_of_votes from the existing features. We also used smooth frequency firing to encode some features, such as the movie's director and stars. Finally, we used a sentence transformer model to turn the overview into a higher dimensional vector. Then we used PCA to reduce the dimensionality of this vector to append it on to the other features.

Citations

We used "all-MiniLM-L6-v2", a sentence transformer model

Disclaimer!!!!!

If running on a device with CUDA, our files (model.py, prep_data.py, etc.) are cuda compatible and WILL use CUDA. You will need to modify Main.py to run with CUDA enabled in order to work with our files in this scenario (because we were not allowed to modify main.py)!!!!!!

Built With

• python
• pytorch