/projects$cat price-prediction-transformer.html

Masked Transformer · Sequential Price Prediction

SLTC · applied R&D · 2025

• Transformers
• PyTorch
• masked modeling
• LSTM
• XGBoost
• MLflow

the idea

Transformers work well in NLP, so I wanted to see whether the same architecture — with the right framing — could compete with classic sequence models on price prediction. I adapted a Transformer with masked modeling and a custom tokenization strategy built for sequential numeric data, rather than reusing one made for words.

testing it properly

I evaluated the model against strong baselines — LSTMs and XGBoost — on the same task, then deployed the best-performing one via MLflow for tracked, reproducible experiments.