See issues open and closed for notes on all fastai work. Issues are used to task work and record solutions throughout the course. Below is a summary of the fastai deep learning experience from December 2018 to May 2019.
This README.md will include all the highlights, links and techytips I have found useful throughout the course.
Created/Taught by Jeremy Howard, Rachel Thomas and team @ USFCA
Link to fastai course homepage
I decided to use Google Cloud Platform compute with £230.58 free credit for 1 year. This is a wonderful tutorial to get started and refer back to. My notes are here
My notes are here
The course is taught in a top down style. Meaning that you learn how something works before you learn why it works. So let's get started.
Recommended and detailed fastai notes by hiromis
Summary:
ImageDataBunch
fastai function. In fastai, everything you model with is going to be a DataBunch object. Basically DataBunch object contains 2 or 3 datasets - it contains your training data, validation data, and optionally test data.size=224
. Models are designed so that the final layer is of size 7 by 7, so we actually want something where if you go 7 times 2 a bunch of times (224 = 7*2^5)learn = create_cnn(data, models.resnet34, metrics=error_rate)
learn.fit_one_cycle(4)
. This number, 4, basically decides how many times do we show the dataset to the model so that it can learn from it. interp.most_confused(min_val=2)
learn.lr_find()
learn.recorder.plot()
learn.unfreeze()
learn.fit_one_cycle(2, max_lr=slice(1e-6,1e-4))
Recommended and detailed fastai notes by hiromis
Summary: