wslでelixir その155

概要

wsl(wsl2じゃない)で、elixirやってみた。
練習問題やってみた。

練習問題

AxonでXORを求よ。

setup

Mix.install(
  [
    {:axon, "~> 0.2.0"},
    {:exla, "~> 0.3.0"},
    {:nx, "~> 0.3.0"},
    {:kino, "~> 0.10.0"}
  ],
  config: [
    nx: [default_backend: EXLA.Backend]
  ]

サンプルコード

x1_input = Axon.input("x1", shape: {nil, 1})
x2_input = Axon.input("x2", shape: {nil, 1})

model =
  x1_input                                     
  |> Axon.concatenate(x2_input)                
  |> Axon.dense(8, activation: :tanh)          
  |> Axon.dense(1, activation: :sigmoid)       



batch_size = 32

data =
  Stream.repeatedly(fn ->
    x1 = Nx.random_uniform({batch_size, 1}, 0, 2)
    x2 = Nx.random_uniform({batch_size, 1}, 0, 2)
    y = Nx.logical_xor(x1, x2)

    {%{"x1" => x1, "x2" => x2}, y}
  end)
 
          
epochs = 10

params =
  model
  |> Axon.Loop.trainer(:binary_cross_entropy, :sgd)
  |> Axon.Loop.run(data, %{}, epochs: epochs, iterations: 1000)

実行結果

Epoch: 0, Batch: 1000, loss: 0.6645805
Epoch: 1, Batch: 1000, loss: 0.6334236
Epoch: 2, Batch: 1000, loss: 0.5894425
Epoch: 3, Batch: 1000, loss: 0.5315313
Epoch: 4, Batch: 1000, loss: 0.4695869
Epoch: 5, Batch: 1000, loss: 0.4142583
Epoch: 6, Batch: 1000, loss: 0.3682923
Epoch: 7, Batch: 1000, loss: 0.3305883
Epoch: 8, Batch: 1000, loss: 0.2995510
Epoch: 9, Batch: 1000, loss: 0.2737042



Axon.predict(model, params, %{
  "x1" => Nx.tensor([[0]]),
  "x2" => Nx.tensor([[1]])
})

Evaluated
#Nx.Tensor<
  f32[1][1]
  EXLA.Backend<host:0, 0.204904100.2404515847.97079>
  [
    [0.9617963433265686]
  ]
>


Axon.predict(model, params, %{
  "x1" => Nx.tensor([[1]]),
  "x2" => Nx.tensor([[0]])
})

Evaluated
#Nx.Tensor<
  f32[1][1]
  EXLA.Backend<host:0, 0.204904100.2404515847.97088>
  [
    [0.9612305164337158]
  ]
>


Axon.predict(model, params, %{
  "x1" => Nx.tensor([[1]]),
  "x2" => Nx.tensor([[1]])
})

Evaluated
#Nx.Tensor<
  f32[1][1]
  EXLA.Backend<host:0, 0.204904100.2404515847.97097>
  [
    [0.05178850144147873]
  ]
>





Axon.predict(model, params, %{
  "x1" => Nx.tensor([[0]]),
  "x2" => Nx.tensor([[0]])
})

Evaluated
#Nx.Tensor<
  f32[1][1]
  EXLA.Backend<host:0, 0.204904100.2404515847.97106>
  [
    [0.01657288148999214]
  ]
>

以上。