Models

class helixnet.models.Sequential(layers_: list[Layer])

A Simple model that propagate through the layers in a linear way

Parameters:

layer (list) – the list which contains the layers

add(layer: Layer) → None

This function can append layers to the model

Parameters:

layer (layer.Layer) – The layer that will be appended to the end of the model

fit(X, Y, loss_func: Callable, optimizer, epochs: int = 1, batch_size: int | None = None, preprocessing: Callable | None = None, metrics: Dict[str, Callable] | None = None)

A high-level training loop with a rich, interactive display.

forward(x: Tensor) → Tensor

Perform a prediction across multiple layers

Args:

x (mg.tensor): the input

Returns:

mg.tensor: the predictions

get_names() → List[str]

Returns a list of layers names

Returns:

list: A list of strings

get_weights() → List[ndarray]

Returns a flat list of all trainable weights in the model.

null_grads() → None

Reset the gradients of every layer

output_shape() → Tuple[int]

A simple function that shows the model’s last layer’s output shape

predict(x: Tensor) → Tensor

This method let the model predict without building computational graph

Parameters:

x (mg.Tensor) – The models input

Return mg.Tensor:

The models predictions

set_weights(weights: List[ndarray])

Sets the model’s weights from a flat list.

Parameters:

weights (List[np.ndarray]) – The weights what will be produced by helixnet.io.load_model()

summary() → None

This method prints the model summary which contains the name of every layer and it’s shape