syna.functions.function module

class syna.functions.function.BroadcastTo(shape: Tuple[int, ...])

Bases: Function

Broadcast x to shape.

backward(gy)

forward(x)

class syna.functions.function.GetItem(slices)

Bases: Function

Supports x[slices] and produces gradient via GetItemGrad.

backward(gy)

forward(x)

class syna.functions.function.GetItemGrad(slices, in_shape)

Bases: Function

Gradient for getitem: scatters gy back into the original shape.

forward(x)

class syna.functions.function.Reshape(shape: Tuple[int, ...])

Bases: Function

Reshape tensor to a given shape.

backward(gy)

forward(x)

class syna.functions.function.Sum(axis, keepdims)

Bases: Function

Sum reduction, supports axis and keepdims.

backward(gy)

forward(x)

class syna.functions.function.SumTo(shape: Tuple[int, ...])

Bases: Function

Sum elements to target shape (inverse of broadcast_to).

backward(gy)

forward(x)

class syna.functions.function.Transpose(axes=None)

Bases: Function

Transpose with optional axes permutation.

backward(gy)

forward(x)

syna.functions.function.broadcast_to(x, shape: Tuple[int, ...]) → Tensor

Broadcast x to the given shape.

syna.functions.function.dropout(x, dropout_ratio=0.5) → Tensor

Dropout during training; identity during evaluation.

syna.functions.function.expand_dims(x, axis: int) → Tensor

Insert a dimension of size 1 at index axis.

syna.functions.function.flatten(x) → Tensor

Flatten all dimensions except the first (batch) dimension.

syna.functions.function.get_item(x, slices) → Tensor

Index into tensor with slices.

syna.functions.function.reshape(x, shape) → Tensor

Reshape tensor; if shape matches returns as_tensor(x).

syna.functions.function.sum(x, axis: Tuple[int, ...] | None = None, keepdims=False) → Tensor

Sum elements along given axes.

syna.functions.function.sum_to(x, shape: Tuple[int, ...]) → Tensor

Sum elements of x so result has shape.

syna.functions.function.transpose(x, axes=None) → Tensor

Transpose tensor along axes.