In this repository we demonstrate the use of various machine learning tools offered by the Elixir ecosystem. Using these tools, we build, train and test a simple convolutional neural network for detecting pneumonia in x-ray chest scans. You can learn more about this repositor in our blog post focused on machine learning with Elixir.
Following Elixir libraries and projects are used:
- Livebook: Elixir equivalent of Jupyter Notebook
- Nx: Multidimensional arrays for Elixir (similar to Python's NumPy)
- EXLA: Client library for Google's XLA compiler
- Axon: Neural networks for Elixir
This guide assumes you already have Elixir installed on your machine. You also need Bazel 3.7.2 to build Tensorflow from source. We recommend installing this specific version with asdf.
- Clone Livebook repo:
$ git clone https://github.com/livebook-dev/livebook.git
$ cd livebook
- Add following dependencies to the
mix.exsfile:
{:exla, "~> 0.1.0-dev", github: "elixir-nx/nx", sparse: "exla"},
{:nx, "~> 0.1.0-dev", github: "elixir-nx/nx", sparse: "nx", override: true},
{:axon, "~> 0.1.0-dev", github: "elixir-nx/axon"}
- Install and compile the dependencies with mix:
$ mix deps.get
$ mix deps.compile
- Clone this repo inside the livebook directory
$ git clone [email protected]:profiq/elixir-ml-example.git
- Start the livebook server
$ mix phx.server
By default, the server will listen on port 4000. You can access it in your browser on [http://localhost:4000].
There you can open the elixir-ml-example/train.livemd notebook. Executing all elixir cells results in
training and storing the model in a file. To test the model you can then
open the elixir-ml-example/test.livemd notebook.
We use the Chest X-Ray Images (Pneumonia) dataset from Kaggle. This repository contains its preprocessed version in a binary format.