I with team ranked 3rd place in the CFT Contest. The task was to identify and correct typos in a person's name. We tried several approaches in this competition, but our two best scores we achieved with a voting of several seq2seq models and lightgbm on linear models predictions. It was an interesting experience primarily because of multitask learning.
prepare data.ipynb- notebook with my data processing code (various counters, TF-IDF and binary features)models.ipynb- notebook with code of learning different models, the quality of which was checked on classification into 2 and 3 classesfinal solution.ipynb- notebook in which I used my best Logistic Regresssion model to solve first task (Full names classification), which was used in the final ensemble, and symspellpy library to solve second task (typos correction)