@spec build_network(map()) :: Axon.t()

Builds the neural network for the Soothsayer model based on the given configuration.

Parameters

• config - A map containing the model configuration.

Returns

An Axon neural network structure.

Examples

iex> config = %{trend: %{enabled: true}, seasonality: %{yearly: %{enabled: true, fourier_terms: 6}}}
iex> network = Soothsayer.Model.build_network(config)
#Axon.Node<...>

@spec display_network(map()) :: Axon.t()

Returns a display-friendly version of the network that outputs a single tensor.

This version can be used with Axon.Display.as_graph/2 since it doesn't use Axon.container with a map output.

Examples

iex> model = Soothsayer.new(config)
iex> input = %{"trend" => Nx.template({1, 1}, :f32), ...}
iex> Axon.Display.as_graph(Soothsayer.Model.display_network(model.config), input)

@spec fit(
  t(),
  %{required(String.t()) => Nx.Tensor.t()},
  Nx.Tensor.t(),
  non_neg_integer()
) :: t()

Fits the Soothsayer model to the provided data.

Parameters

• model - A Soothsayer.Model struct.
• x - A map of input tensors.
• y - A tensor of target values.
• epochs - The number of training epochs.

Returns

An updated Soothsayer.Model struct with fitted parameters.

Examples

iex> model = Soothsayer.Model.new(config)
iex> x = %{"trend" => trend_tensor, "yearly" => yearly_tensor, "weekly" => weekly_tensor}
iex> y = target_tensor
iex> fitted_model = Soothsayer.Model.fit(model, x, y, 100)
%Soothsayer.Model{...}

@spec new(map()) :: t()

Creates a new Soothsayer.Model struct with the given configuration.

Parameters

• config - A map containing the model configuration.

Returns

A new Soothsayer.Model struct.

Examples

iex> config = %{trend: %{enabled: true}, seasonality: %{yearly: %{enabled: true, fourier_terms: 6}}}
iex> Soothsayer.Model.new(config)
%Soothsayer.Model{network: ..., params: nil, config: ^config}

@spec predict(t(), %{required(String.t()) => Nx.Tensor.t()}) :: %{
  combined: Nx.Tensor.t(),
  trend: Nx.Tensor.t(),
  yearly_seasonality: Nx.Tensor.t(),
  weekly_seasonality: Nx.Tensor.t()
}

Makes predictions using a fitted Soothsayer model.

Parameters

• model - A fitted Soothsayer.Model struct.
• x - A map of input tensors.

Returns

A map containing the predicted values for each component and the combined prediction.

Examples

iex> fitted_model = Soothsayer.Model.fit(model, training_x, training_y, 100)
iex> x = %{"trend" => future_trend_tensor, "yearly" => future_yearly_tensor, "weekly" => future_weekly_tensor}
iex> predictions = Soothsayer.Model.predict(fitted_model, x)
%{
  combined: #Nx.Tensor<...>,
  trend: #Nx.Tensor<...>,
  yearly_seasonality: #Nx.Tensor<...>,
  weekly_seasonality: #Nx.Tensor<...>
}