This is submission of the challenge of the Fellowship.AI 2019, for which I took the ULMFiT Sentiment problem.
The most important sections are:
- Part 1: Language Model and Fine Tuning
- Part 2: Text Classification Task with ULMFiT
- The discussion at the end of part 2, with some ideas of weaknesses and possible improvements.
But to have a complete picture:
- Part 0 deals with exploring the data set and trying different feature extraction approaches
- Part 3 is a "bonus" with multi-class classification (including the neutral cases)
(c) César Fuentes, 2019.