Titanic with Explorer

Mix.install([
  {:exla, "~> 0.6"},
  {:axon, "~> 0.5"},
  {:kino, "~> 0.12"},
  {:kino_vega_lite, "~> 0.1"},
  {:explorer, "~> 0.8"}
])

エイリアス

alias Explorer.DataFrame
alias Explorer.Series
require Explorer.DataFrame

データの準備

Kaggle のタイタニックデータをダウンロードする

https://www.kaggle.com/competitions/titanic/data

train_data_input = Kino.Input.file("train data")

test_data_input = Kino.Input.file("test data")

前処理

train_data =
  train_data_input
  |> Kino.Input.read()
  |> Map.get(:file_ref)
  |> Kino.Input.file_path()
  |> DataFrame.from_csv!()

Kino.DataTable.new(train_data)

id_list = Series.to_list(train_data["PassengerId"])

label_tensor =
  train_data["Survived"]
  |> Series.to_tensor(backend: EXLA.Backend)
  |> Nx.as_type(:f32)
  |> Nx.new_axis(1)
  |> Nx.to_batched(1)
  |> Enum.to_list()

inputs = DataFrame.discard(train_data, ["PassengerId", "Survived"])

Kino.DataTable.new(inputs)

dropped_inputs = DataFrame.discard(inputs, ["Cabin", "Name", "Ticket"])

Kino.DataTable.new(dropped_inputs)

filled_inputs =
  dropped_inputs
  |> DataFrame.put("Age", Series.fill_missing(train_data["Age"], :mean))
  |> DataFrame.put("Embarked", Series.fill_missing(train_data["Embarked"], "S"))

Kino.DataTable.new(filled_inputs)

dummied_inputs =
  filled_inputs
  |> DataFrame.dummies(["Sex", "Embarked"])
  |> DataFrame.concat_columns(DataFrame.discard(filled_inputs, ["Sex", "Embarked"]))

Kino.DataTable.new(dummied_inputs)

input_tensor_list =
  dummied_inputs
  |> DataFrame.to_columns()
  |> Map.values()
  |> Nx.tensor(backend: EXLA.Backend)
  |> Nx.transpose()
  |> Nx.to_batched(1)
  |> Enum.to_list()

defmodule PreProcess do
  def load_csv(kino_input) do
    kino_input
    |> Kino.Input.read()
    |> Map.get(:file_ref)
    |> Kino.Input.file_path()
    |> DataFrame.from_csv!()
  end

  def fill_empty(data, fill_map) do
    fill_map
    |> Enum.reduce(data, fn {column_name, fill_value}, acc ->
      DataFrame.put(
        acc,
        column_name,
        Series.fill_missing(data[column_name], fill_value)
      )
    end)
  end

  def replace_dummy(data, columns_names) do
    data
    |> DataFrame.dummies(columns_names)
    |> DataFrame.concat_columns(DataFrame.discard(data, columns_names))
  end

  def to_tensor(data) do
    data
    |> DataFrame.to_columns()
    |> Map.values()
    |> Nx.tensor(backend: EXLA.Backend)
    |> Nx.transpose()
    |> Nx.to_batched(1)
    |> Enum.to_list()
  end

  def process(kino_input, id_key, label_key) do
    data_org = load_csv(kino_input)

    id_list = Series.to_list(data_org[id_key])

    has_label_key =
      data_org
      |> DataFrame.names()
      |> Enum.member?(label_key)

    label_list =
      if has_label_key do
        data_org[label_key]
        |> Series.to_tensor(backend: EXLA.Backend)
        |> Nx.as_type(:f32)
        |> Nx.new_axis(1)
        |> Nx.to_batched(1)
        |> Enum.to_list()
      else
        nil
      end

    inputs =
      if has_label_key do
        DataFrame.discard(data_org, [id_key, label_key])
      else
        DataFrame.discard(data_org, [id_key])
      end
      |> DataFrame.discard(["Cabin", "Name", "Ticket"])
      |> fill_empty(%{"Age" => :mean, "Embarked" => "S", "Fare" => :mean})
      |> replace_dummy(["Sex", "Embarked"])
      |> to_tensor()

    {id_list, label_list, inputs}
  end
end

{
  train_id_list,
  train_label_list,
  train_inputs
} = PreProcess.process(train_data_input, "PassengerId", "Survived")

{
  test_id_list,
  test_label_list,
  test_inputs
} = PreProcess.process(test_data_input, "PassengerId", "Survived")

モデルの定義

model =
  Axon.input("input", shape: {nil, 10})
  |> Axon.dense(48, activation: :tanh)
  |> Axon.dropout(rate: 0.2)
  |> Axon.dense(48, activation: :tanh)
  |> Axon.dense(1, activation: :sigmoid)

学習

train_data = Enum.zip(train_inputs, train_label_list)

loss_plot =
  VegaLite.new(width: 300)
  |> VegaLite.mark(:line)
  |> VegaLite.encode_field(:x, "step", type: :quantitative)
  |> VegaLite.encode_field(:y, "loss", type: :quantitative)
  |> Kino.VegaLite.new()

acc_plot =
  VegaLite.new(width: 300)
  |> VegaLite.mark(:line)
  |> VegaLite.encode_field(:x, "step", type: :quantitative)
  |> VegaLite.encode_field(:y, "accuracy", type: :quantitative)
  |> Kino.VegaLite.new()

Kino.Layout.grid([loss_plot, acc_plot], columns: 2)

trained_state =
  model
  |> Axon.Loop.trainer(:mean_squared_error, Axon.Optimizers.adam(0.0005))
  |> Axon.Loop.metric(:accuracy, "accuracy")
  |> Axon.Loop.kino_vega_lite_plot(loss_plot, "loss", event: :epoch_completed)
  |> Axon.Loop.kino_vega_lite_plot(acc_plot, "accuracy", event: :epoch_completed)
  |> Axon.Loop.run(train_data, %{}, epochs: 50, compiler: EXLA)

未知データに対する推論

results =
  test_inputs
  |> Nx.concatenate()
  |> then(&Axon.predict(model, trained_state, &1))
  |> Nx.to_flat_list()
  |> Enum.map(&round(&1))
  |> then(
    &%{
      "PassengerId" => test_id_list,
      "Survived" => &1
    }
  )
  |> DataFrame.new()

Kino.DataTable.new(results)

results
|> DataFrame.dump_csv!()
|> then(&Kino.Download.new(fn -> &1 end, filename: "result.csv"))