snnlab.lang.ops

Complete declared API of the ops module, with signatures, data fields, validation and source.

Small serialisable operation vocabulary.

The signatures, defaults, fields, docstrings and implementation excerpts below are generated from the Python source. Annotations are shown as declared; unannotated means the source supplies no type annotation. These pages document callable surfaces, including legacy support utilities, without promising backend support for every declaration.

Symbol	Kind
linear	function
reduce	function
divide	function

linear

View source

def linear(source: Signal, *, size: int, name: str, trainable: bool=True) -> Signal

Author a linear operation with a zero-initialized matrix parameter. Its output preserves leading dimensions and replaces the final feature dimension with size.

Parameter	Annotation	Default	Meaning
`source`	`Signal`	required	Defined by the source contract and implementation below.
`size`	`int`	required	Defined by the source contract and implementation below.
`name`	`str`	required	Name used to identify the authored or rendered object.
`trainable`	`bool`	`True`	Defined by the source contract and implementation below.

Return annotation: Signal.

Return expressions (branch-dependent; names refer to the linked implementation):

net.operation('linear', source, name=name, shape=(*source.shape[:-1], size), unit=source.unit, signal_type='continuous', parameters=(weight,), size=size, trainable=trainable)

Implementation

def linear(source: Signal, *, size: int, name: str, trainable: bool = True) -> Signal:
    net = source.network
    weight = net.parameter(
        f"{name}.weight",
        shape=(size, source.shape[-1]),
        initializer=Constant(0.0),
        constraint=None,
    )
    return net.operation(
        "linear",
        source,
        name=name,
        shape=(*source.shape[:-1], size),
        unit=source.unit,
        signal_type="continuous",
        parameters=(weight,),
        size=size,
        trainable=trainable,
    )

reduce

View source

def reduce(source: Signal, *, operation: str, over: str, name: str, window: str='full', mask: Signal | None=None) -> Signal

Author a reduction over the explicit time axis, removing that axis from the output shape. operation selects the reduction kind; mask and window are recorded for execution.

Parameter	Annotation	Default	Meaning
`source`	`Signal`	required	Defined by the source contract and implementation below.
`operation`	`str`	required	Defined by the source contract and implementation below.
`over`	`str`	required	Defined by the source contract and implementation below.
`name`	`str`	required	Name used to identify the authored or rendered object.
`window`	`str`	`'full'`	Defined by the source contract and implementation below.
`mask`	`Signal \| None`	`None`	Defined by the source contract and implementation below.

Return annotation: Signal.

Return expressions (branch-dependent; names refer to the linked implementation):

source.network.operation(f'reduce_{operation}', sources, name=name, shape=tuple((dimension for index, dimension in enumerate(source.shape) if index != time_axis)), unit=source.unit, window=window, mask=mask.id if mask else None)

Explicit exceptions in this implementation; called helpers may raise additional errors:

Explicit exception expression
`ValueError('v1 reductions support only an explicit time axis')`

Implementation

def reduce(
    source: Signal,
    *,
    operation: str,
    over: str,
    name: str,
    window: str = "full",
    mask: Signal | None = None,
) -> Signal:
    if over != "time" or "time" not in source.shape:
        raise ValueError("v1 reductions support only an explicit time axis")
    time_axis = source.shape.index("time")
    sources = [source] + ([mask] if mask else [])
    return source.network.operation(
        f"reduce_{operation}",
        sources,
        name=name,
        shape=tuple(
            dimension
            for index, dimension in enumerate(source.shape)
            if index != time_axis
        ),
        unit=source.unit,
        window=window,
        mask=mask.id if mask else None,
    )

divide

View source

def divide(source: Signal, denominator: Signal, *, name: str, unit: str) -> Signal

Author elementwise division using source and denominator, retaining source shape and declaring the supplied output unit.

Parameter	Annotation	Default	Meaning
`source`	`Signal`	required	Defined by the source contract and implementation below.
`denominator`	`Signal`	required	Defined by the source contract and implementation below.
`name`	`str`	required	Name used to identify the authored or rendered object.
`unit`	`str`	required	Declared physical unit; projection weights use uS.

Return annotation: Signal.

Return expressions (branch-dependent; names refer to the linked implementation):

source.network.operation('divide', [source, denominator], name=name, shape=source.shape, unit=unit)

Implementation

def divide(source: Signal, denominator: Signal, *, name: str, unit: str) -> Signal:
    return source.network.operation(
        "divide",
        [source, denominator],
        name=name,
        shape=source.shape,
        unit=unit,
    )

snnlab.lang.ops

linearCopy Anchor Link

reduceCopy Anchor Link

divideCopy Anchor Link

On this page

linear

reduce

divide