AIRush2

Project Outline

2019 Naver AI hackathon - CTR Prediction

Requirements

• Python 3.x
• Pytorch 1.0
• NSML

Model architecture

• The code is implemented based on baseline model (FCN)
• Binary Classification Model
• CNN + Embedding -> FCN
• Apply Resnet 18 to extract the image features and target encoding method to embed the sequence data (user reading history)
• Large batch size to increase the training efficiency by using shallow FCN model structure
• The final model is an ensemble model that uses hard voting method for N different models