Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Update "Horses or humans" example. #543

Merged
merged 1 commit into from
Nov 14, 2023
Merged

Update "Horses or humans" example. #543

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 11 additions & 11 deletions notebooks/vision/horses_or_humans.livemd
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,12 +2,12 @@

```elixir
Mix.install([
{:axon, "~> 0.3.0"},
{:nx, "~> 0.4.0", sparse: "nx", override: true},
{:exla, "~> 0.4.0", sparse: "exla", override: true},
{:stb_image, "~> 0.5.2"},
{:req, "~> 0.3.1"},
{:kino, "~> 0.7.0"}
{:axon, "~> 0.6.0"},
{:nx, "~> 0.6.0"},
{:exla, "~> 0.6.0"},
{:stb_image, "~> 0.6.0"},
{:req, "~> 0.4.5"},
{:kino, "~> 0.11.0"}
])

Nx.global_default_backend(EXLA.Backend)
Expand All @@ -24,7 +24,7 @@ We will be using the [Horses or Humans Dataset](https://laurencemoroney.com/data

```elixir
%{body: files} =
Req.get!("https://storage.googleapis.com/laurencemoroney-blog.appspot.com/horse-or-human.zip")
Req.get!("https://storage.googleapis.com/learning-datasets/horse-or-human.zip")

files = for {name, binary} <- files, do: {List.to_string(name), binary}
```
Expand Down Expand Up @@ -121,7 +121,7 @@ defmodule HorsesHumans.DataProcessing do
images = images / 255.0

# Optional vertical/horizontal flip
u = Nx.random_uniform({})
{ u, _new_key } = Nx.Random.key(1987) |> Nx.Random.uniform()

cond do
u < 0.25 -> images
Expand Down Expand Up @@ -203,7 +203,7 @@ optimizer = Polaris.Optimizers.adam(learning_rate: 1.0e-4)

params =
model
|> Axon.Loop.trainer(:categorical_cross_entropy, optimizer, :identity, log: 1)
|> Axon.Loop.trainer(:categorical_cross_entropy, optimizer, log: 1)
|> Axon.Loop.metric(:accuracy)
|> Axon.Loop.run(data, %{}, epochs: 10, iterations: batches_per_epoch)
```
Expand All @@ -218,7 +218,7 @@ We can improve the training by applying gradient centralization. It is a techniq
centralized_optimizer = Polaris.Updates.compose(Polaris.Updates.centralize(), optimizer)

model
|> Axon.Loop.trainer(:categorical_cross_entropy, centralized_optimizer, :identity, log: 1)
|> Axon.Loop.trainer(:categorical_cross_entropy, centralized_optimizer, log: 1)
|> Axon.Loop.metric(:accuracy)
|> Axon.Loop.run(data, %{}, epochs: 10, iterations: batches_per_epoch)
```
Expand Down Expand Up @@ -246,4 +246,4 @@ _Note: the model output refers to the probability that the image presents a hors

<!-- livebook:{"break_markdown":true} -->

The website from where we loaded the dataset also includes a validation set, in case you want to experiment further!
You can find a validation set [here](https://storage.googleapis.com/learning-datasets/validation-horse-or-human.zip), in case you want to experiment further!
Loading