Key points
Lesson four
Drop out
fastai proc_df()
what we do with continuos data 1:40min
embedding matrix - distributed representation
you can pass in custom matrics in your .fit()
method as .fit(lr,3,metrics=[exp_rmspe]
exp_rmspe was crate earlier in the notebook. Does'nt not chabge training on what is printed out.
Lesson 5
dictionary mapping
fit function 57min
bias
broadcasting 1:05
capital F thing
Stochastic gradient descent
momentum
admam
weight decay l2 reg etc
linear interpolation
chain rule
Lesson 6
questions
learner.fit vs fit(model)