NN Components

Module: Neural Network Module Logic

Classes

Class	Description
Module	Generic neural network module.
Flatten	Flatten layer implementation.
Linear	Linear layer implementation.
AvgPool2d	2D-Average pooling layer implementation.
DotProductSimilarity	Dot product similarity module.
ReLU	ReLU layer implementation.
Conv2d	Conv2D layer implementation.
Parameter	Parameter class.
LinearRegression	Linear regression implementation
LogisticRegression	Logistic regression implementation
Prophet	Prophet forecasting implementation

Model Clients

Class	Description
ProphetClient	ModelClient for Prophet models
SklearnClient	ModelClient for Scikit-learn models
TorchClient	ModelClient for PyTorch models
ModelClientMeta	ML model client metaclass
ModelClient	ML model client

Class: Module

Method	Description
forward(x: na.NadaArray) -> na.NadaArray	Abstract method for forward pass.
__call__(*args, **kwargs) -> na.NadaArray	Proxy for forward pass.
__named_parameters(prefix: str) -> Iterator[Tuple[str, Parameter]]	Recursively generates all parameters in the module.
named_parameters() -> Iterator[Tuple[str, Parameter]]	Generates all parameters in the module.
__numel() -> Iterator[int]	Recursively generates number of elements in each parameter.
numel() -> int	Returns total number of elements in the module.
load_state_from_network(name: str, party: Party, nada_type: NadaInteger) -> None	Loads the model state from the Nillion network.

Class: Flatten

Method	Description
__init__(start_dim: int = 1, end_dim: int = -1) -> None	Initializes the flatten layer with start and end dimensions.
forward(x: na.NadaArray) -> na.NadaArray	Forward pass. Flattens the input tensor.

Class: Linear

Method	Description
__init__(in_features: int, out_features: int, include_bias: bool = True) -> None	Initializes the linear layer with input features, output features, and an optional bias.
forward(x: na.NadaArray) -> na.NadaArray	Forward pass. Applies a linear transformation to the input.

Class: AvgPool2d

Method	Description
__init__(kernel_size: ShapeLike2d, stride: Optional[ShapeLike2d] = None, padding: ShapeLike2d = 0) -> None	Initializes the 2D average pooling layer.
forward(x: na.NadaArray) -> na.NadaArray	Forward pass. Applies average pooling to the input.

Class: DotProductSimilarity

Method	Description
forward(x_1: na.NadaArray, x_2: na.NadaArray) -> na.NadaArray	Forward pass. Computes the dot product similarity between two input arrays.

Class: ReLU

Method	Description
forward(x: na.NadaArray) -> na.NadaArray	Forward pass. Applies the ReLU activation function to the input.
static _rational_relu(value: Union[na.Rational, na.SecretRational]) -> Union[na.Rational, na.SecretRational]	Element-wise ReLU logic for rational values.
static _relu(value: NadaType) -> Union[PublicInteger, SecretInteger]	Element-wise ReLU logic for NadaType values.

Class: Conv2d

Method	Description
__init__(in_channels: int, out_channels: int, kernel_size: ShapeLike2d, padding: ShapeLike2d = 0, stride: ShapeLike2d = 1, include_bias: bool = True) -> None	Initializes the 2D convolutional layer.
forward(x: na.NadaArray) -> na.NadaArray	Forward pass. Applies 2D convolution to the input.

Class: Parameter

Method	Description
__init__(shape: ShapeLike = 1) -> None	Initializes the parameter with a given shape.
numel() -> int	Returns the number of elements in the parameter.
load_state(state: na.NadaArray) -> None	Loads a provided NadaArray as the new parameter state.

Linear Models

Class Name	Description	Methods
LinearRegression	Linear regression implementation	__init__(in_features: int, include_bias: bool = True) -> None forward(x: na.NadaArray) -> na.NadaArray
LogisticRegression	Logistic regression implementation	__init__(in_features: int, out_features: int, include_bias: bool = True) -> None forward(x: na.NadaArray) -> na.NadaArray

Time Series Models

Class Name	Description	Methods
Prophet	Prophet forecasting implementation	__init__(n_changepoints: int, growth: str = "linear", yearly_seasonality: bool = True, weekly_seasonality: bool = True, daily_seasonality: bool = False, seasonality_mode: str = "additive") -> None predict(dates: np.ndarray, floor: na.NadaArray, t: na.NadaArray) -> na.NadaArray predict_trend(floor: na.NadaArray, t: na.NadaArray) -> na.NadaArray predict_seasonal_comps(dates: np.ndarray) -> Tuple[na.NadaArray, na.NadaArray] make_seasonality_features(dates: np.ndarray, seasonalities: Dict[str, Any]) -> Dict[str, na.NadaArray] ensure_numeric_dates(dates: np.ndarray) -> np.ndarray __call__(dates: np.ndarray, floor: na.NadaArray, t: na.NadaArray) -> na.NadaArray forward(dates: np.ndarray, floor: na.NadaArray, t: na.NadaArray) -> na.NadaArray

Model Clients

Class Name	Description	Methods
ProphetClient	ModelClient for Prophet models	__init__(model: prophet.forecaster.Prophet) -> None
SklearnClient	ModelClient for Scikit-learn models	__init__(model: sklearn.base.BaseEstimator) -> None
TorchClient	ModelClient for PyTorch models	__init__(model: nn.Module) -> None
ModelClientMeta	ML model client metaclass	__call__(cls, *args, **kwargs) -> object
ModelClient	ML model client	export_state_as_secrets(name: str, nada_type: NillionType) -> Dict[str, NillionType] __ensure_numpy(array_like: Any) -> np.ndarray

For more examples, please visit our Github Repository Examples.