Function encode

pub fn encode<T: FloatExt, S: Data<Elem = T>, D: Dimension, B: AutodiffBackend<FloatElem = T>>(
    device: &B::Device,
    data: ArrayBase<S, D>,
    fourier_features: NonZeroUsize,
    fourier_scale: Positive<f64>,
    num_blocks: NonZeroUsize,
    learning_rate: Positive<f64>,
    num_epochs: usize,
    mini_batch_size: Option<NonZeroUsize>,
    seed: u64,
) -> Result<Array<u8, Ix1>, FourierNetworkCodecError>

Encodes the data by training a fourier feature neural network.

The fourier_features are randomly sampled from a normal distribution with zero mean and fourier_scale standard deviation.

The neural network consists of num_blocks blocks.

The network is trained for num_epochs using the learning_rate and mini-batches of mini_batch_size if mini-batching is enabled.

All random numbers are generated using the provided seed.

Errors

Errors with

• FourierNetworkCodecError::NonFiniteData if any data element is non-finite (infinite or NaN)
• FourierNetworkCodecError::NeuralNetworkError if an error occurs during the neural network computation