tensorplay.autograd.forward_ad API

Enters a new forward grad level and returns its index.

This level can be used to make and unpack dual Tensors to compute forward gradients. Levels cannot nest: entering while another level is active raises, so all forward AD computation happens inside one level.

Exits a forward grad level.

This deletes all the tangents associated with this level. Only exiting the most recently entered level is allowed.

Associates a tensor value with its tangent to create a “dual tensor”.

The result is a new tensor aliased to tensor with tangent attached as its forward gradient. The tangent can be recovered with unpack_dual().

Given a function f whose jacobian is J, this computes the Jacobian-vector product (jvp) between J and a given vector v:

>>> # xdoctest: +SKIP("Undefined variables")
>>> with dual_level():
...     inp = make_dual(x, v)
...     out = f(inp)
...     y, jvp = unpack_dual(out)

Unpacks a dual tensor into its primal value and forward gradient.

Returns (primal, tangent) where primal is a view of tensor’s primal and tangent is tensor’s tangent (None when tensor carries no tangent at level).