description: Namedtuple of quantities that may be traced from tfp.math.minimize
.
View source on GitHub |
Namedtuple of quantities that may be traced from tfp.math.minimize
.
tfp.math.MinimizeTraceableQuantities(
step, loss, gradients, parameters, has_converged, convergence_criterion_state
)
These are (in order):
step
: int Tensor
index (starting from zero) of the current optimization
step.loss
: float Tensor
value returned from the user-provided loss_fn
.gradients
: list of Tensor
gradients of loss
with respect to the
parameters.parameters
: list of Tensor
values of parameters being optimized. This
corresponds to the trainable_variables
passed in to minimize
.has_converged
: boolean Tensor
of the same shape as loss_fn
, with
True
values corresponding to loss entries that have converged according
to the user-provided convergence criterion. If no convergence criterion
was specified, this is None
.convergence_criterion_state
: structure of Tensor
s containing any
auxiliary state (e.g., moving averages of loss or other quantities)
maintained by the user-provided convergence criterion.Attributes | |
---|---|
step
|
|
loss
|
|
gradients
|
|
parameters
|
|
has_converged
|
|
convergence_criterion_state
|