dipas.optimize module¶
-
class
dipas.optimize.
JacobianAdapter
(f_compute, *, ref_data, verbose=False, history=False)¶ Bases:
object
This class allows combination with external optimizers which require the Jacobian.
Can be used together with
scipy.optimize.least_squares
for example.- Parameters
f_compute (callable) – This function should compute the desired quantity, given a tensor of inputs. It will be called with a single argument, a tensor of shape
(N,)
and must output a single tensor of shape(M,)
.ref_data (torch.Tensor, shape (M,)) – The reference data to compute the residuals w.r.t. the output of f_compute. Must have the same shape as the output of f_compute.
verbose (bool, optional) – If True then at every iteration the current mean squared error is printed to
sys.stdout
.history (bool, optional) – If True then at every iteration the current estimate and residuals are saved in the history attribute.
-
step
¶ The current step during the optimization. This attribute is incremented by one for each
__call__
of the adapter.- Type
int
-
history
¶ If the history parameter is set to true then this list is appended the current parameter estimate and residuals as a tuple on every iteration.
- Type
list
-
class
Progress
(estimate, residual)¶ Bases:
tuple
-
property
estimate
¶ Alias for field number 0
-
property
residual
¶ Alias for field number 1
-
property
-
jacobian
(_)¶ Return the Jacobian corresponding to the last estimate.