Implementing kernel
C++ engine symbols that back this Python API.Return the running minimum of input along dim.
Output i along dim is the minimum of elements 0..i. Only the
values come back; the reference framework also returns the index of
each running minimum, which is not computed here.
Parameters
inputTensorSource tensor.
dimint= -1Dimension to scan along.
Returns
TensorSame shape and dtype as input.
Notes
The gradient of output i flows to the element that set the running
minimum at step i; differentiable twice.
Examples
>>> import lucid
>>> x = lucid.tensor([1.0, 3.0, 2.0, 4.0])
>>> lucid.cummin(x, dim=0)
Tensor([1., 1., 1., 1.])