Thanks @AstonJ

Do you think (with your work) it’s a good book for Elixir users? Is it an easy read?

It is definitely a good and well written book, with a practical approach and only a couple of “heavy-math” chapters. In general I liked it and I found it easy to follow, even if I didn’t have much experience with numpy. In the first chapters I struggled a bit to replicate the examples in Nx, but mainly because I was completely new to the library and to the concepts, that’s part of the game.

If so maybe we should run a book club on it

I think it can be a valuable reading if you wanna start with ML, on the other hand, the book is 3 years old and given the pace of innovation in ML, maybe there are more up to date resources out there The basis won’t change tho’, but for example I’d have like some chapters on more advanced topics such as NLP, transformers, reinforcement learning (but I guess they deserve a book on their own ).

Paolo Perrotta is a great author - he wrote one of my favourite Ruby books: Metaprpogramming Ruby!

I never read it, but I saw a talk on Git by him once and it was great

Hi Nick,

Thank you for doing this. It is really great.

I started reading the book, and follow along with your livebooks. I have only just started so I am only at Chapter two at the moment.

By evaluating your code for the Linear Regression example, I noticed that it always goes through all the iterations, whereas the Python code in the book stops when the error is no longer smaller then the previous error.

By using a reduce_while instead of a reduce in the train function, you could mimic that behaviour. Here is how I changed it:

def train(x, y, iterations, lr) when is_list(x) and is_list(y) do
    Enum.reduce_while(0..iterations, 0, fn i, w ->
      current_loss = loss(x, y, w)

      IO.puts("Iteration #{i} => Loss: #{current_loss}")

      cond do
        loss(x, y, w + lr) < current_loss -> {:cont, w + lr}
        loss(x, y, w - lr) < current_loss -> {:cont, w - lr}
        true -> {:halt, w}
      end
    end)
  end

It will then stop at the 184th iteration, just like in the book

Regards,

Hein

Nice one Nick! Do you think (with your work) it’s a good book for Elixir users? Is it an easy read? If so maybe we should run a book club on it

Paolo Perrotta is a great author - he wrote one of my favourite Ruby books: Metaprpogramming Ruby! Here’s my recollection of it in his spotlight on DT:

@NickGnd This is so timely & amazing. Thank you.

I’ve tried the Dockyard tutorials at first since I didn’t want to leave Elixir-land but soon found it easier to understand nx/axon it after I understood the analogous libraries in Python.

Be happy to provide feedback on the livebooks as I progress.

Have a great day too

Hey man. I’ve tried ML before that and it looks like this book was the only one that made sense to me. Other were a lot more complicated and math heavy.

Learning Elixir now and I am so happy that you created such repo. Amazing job!