Deep Learning na Unb (Brasília) - Parte 1 - Lição 1

[pierreguillou]

[ EDIT 24/10/2019 ]: para baixar um dataset de uma competição Kaggle, a gente precisa cadastrar-se no site, aceitar as regras da competição e usar a API Kaggle. No início deste notebook lição3-planeta.ipynb, há todas as informações e códigos necessários por isso.

---

Para aqueles que desejam criar um classificador de imagens até a próxima semana, usando o notebook lesson1-pets.ipynb da lição 1 e datasets do Kaggle, veja este post no medium: “fast.ai: How I built a deep learning application to detect invasive species in just 1 day (and for $12.60)”.

E mais um post sobre o Quick_Draw competition que fala do uso do Kaggle para começar a criar um classificador (thanks @EricPB):

@EricPB

For those willing to try their first Kaggle competition, or those with little experience like me, what @radek created on Kaggle and GitHub for this Quick_Draw competition is pure gold:
He takes you through the initial steps of downloading the right data,
how to process/transform it to fit into FastaiV1,
how/where to modify the tuning (image_size, full or partial dataset_size, number of epochs),
and even include the code to optimize it (TTA) then generate a submission

It’s truly an amazing package for an on-going competition.

Of course, his starter pack “as it is” won’t get you a medal.
But if you dive in and figure out how to play with FastaiV1 options/parameters, you have a unique chance to compete and see your Public Leaderboard score move as you explore (remember: only 5 submissions per day).