Interim continuous-adjoint shooting implementation #885
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…pute_partials need to be computed
…ding parameter, control, and polynomial control
…id muddying the line between distributions and transcriptions. DirectShootingPhase is solving the brachistochrone problem when the output grid is the same as the input grid, but fails if they are different.
…thin their respective control groups.
…n tests that are covered by the DirectShootingPhase tests.
robfalck
changed the title
kaushikponnapalli
approved these changes
Jan 5, 2023
johnjasa
approved these changes
Jan 5, 2023
dymos/phase/direct_shooting_phase.py
Outdated
| self.options.declare('propagate_derivs', types=bool, default=True, | ||
| desc='If True, propagate the state and derivatives of the state and time with respect to ' | ||
| 'the integration parameters. If False, only propagate the primal states. If only ' | ||
| 'using this transcription to propagate an ODE and derivatives are needed, setting ' |
There was a problem hiding this comment.
maybe it should be "and derivatives are not needed"? Might be missing a word or I'm misunderstanding
| If a string, the file to which the result of the simulation will be saved. | ||
| If None, no record of the simulation will be saved. | ||
|
|
||
| Returns |
There was a problem hiding this comment.
Minor detail: does this return problem if an error is just raised? I'm not sure if this docstring should be different
There was a problem hiding this comment.
Understood, the NotImplementedError should be overridden ultimately.
|
|
||
| Returns | ||
| ------- | ||
| problem |
There was a problem hiding this comment.
Same minor detail comment about the return docstring as above
dymos/transcriptions/grid_data.py
Outdated
| """ | ||
| Provides a dict of node info and locations for a uniformly distributed set of n nodes on the range [-1, 1]. | ||
|
|
||
| This distribution is not to be used to define polynomials, since equally space nodes |
| self.options.declare('propagate_derivs', types=bool, default=True, | ||
| desc='If True, propagate the state and derivatives of the state and time with respect to ' | ||
| 'the integration parameters. If False, only propagate the primal states. If only ' | ||
| 'using this transcription to propagate an ODE and derivatives are needed, setting ' |
| elif var_type == 'control_rate': | ||
| control_name = var[:-5] | ||
| path = f'control_rates:{control_name}_rate' | ||
| control_name = var[:-5] |
There was a problem hiding this comment.
second control_name definition unnecessary?
|
|
||
| def setup(self): | ||
| """ | ||
| Add the necessary I/O and storage for the RKIntegrationComp. |
There was a problem hiding this comment.
RKIntegrationComp -> ODEIntegrationComp
|
|
||
| return f, f_x, f_t, f_theta | ||
|
|
||
| def _f_augmented(self, t, y, theta, dtheta_dz): |
There was a problem hiding this comment.
You know, this is just really neat stuff!
| A component which interpolates control values in 1D using Vandermonde interpolation. | ||
|
|
||
| Takes training values for control variables at given _input_ nodes, | ||
| broadcaasts them to _discretization_ nodes, and then interpolates the discretization values |
|
|
||
| def compute_partials(self, inputs, partials, discrete_inputs=None): | ||
| """ | ||
| Compute derivatives interpolated control values and rates wrt the inputs. |
There was a problem hiding this comment.
"derivatives" -> "derivatives of"?
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
This is an interim implementation of a continuous-adjoint direct shooting implementation for Dymos.
Ultimately, this implementation will replace the current ExplicitShooting transcription but that will come in a following PR since this one is already massive.
DirectShootingPhase, a temporary name that will ultimately beExplicitShootingPhase.AnalyticPhase,DirectShootingPhaseautomatically assumes a particular transcription (DirectShooting, which will be renamedExplicitShootingin a future PR).DirectShootinghas a few key differences with the existingExplicitShootingimplementation.SolveIVPimplementation in a future PR, making for easier maintenance.The Concept of Grids
Dymos currently conflates the notion of Grids and Transcriptions a bit.
With this DirectShootingPhase, the points at which output are desired do not necessarily line up with the grid on which the control variables are defined.
Allowing the user to specify a whole transcription object at which outputs are desired seems overkill, so this PR introduces
GaussLobattoGrid,RadauGrid, andUniformGrid. These define node placement without all of the extra setup bloat.In a future PR, these will be used to specify output points for simulate and timeseries output points.
Some rework was done on
ControlGroupandPolynomialControlGroupto allow control outputs to be provided on different grids, with the requirement that the segment spacing in the grids remains the same (otherwise control rate interpolation would be nonsense).Other things handled along the way.
pep8dependency withpycodestyle. Note there was a significant amount of reformatting to satisfy pycodestyle.Related Issues
Backwards incompatibilities
None
New Dependencies
Replaced pep8 testing dependency with pycodestyle.