HybridNonlinearFactor Relinearization #1831
Conversation
…d arbitrary scalars
|
@dellaert take a look while I finish adding some tests for |
| for (auto &&f : motion_models) { | ||
| components.push_back(std::dynamic_pointer_cast<NonlinearFactor>(f)); | ||
| components.push_back( | ||
| {std::dynamic_pointer_cast<NonlinearFactor>(f), 0.0}); |
There was a problem hiding this comment.
@dellaert here's one reason to keep the vector based constructor. Allows us to define the tree incrementally like this.
…same as the discrete key cardinality
|
@dellaert based on our discussion, I have updated the API and tests. |
| @@ -107,25 +107,13 @@ class GTSAM_EXPORT HybridGaussianConditional | |||
| const Conditionals &conditionals); | |||
|
|
|||
| /** | |||
| * @brief Make a Gaussian Mixture from a list of Gaussian conditionals | |||
| * @brief Make a Gaussian Mixture from a vector of Gaussian conditionals. | |||
| * The DecisionTree-based constructor is preferred over this one. | |||
There was a problem hiding this comment.
Kept this constructor for the python wrapper, but added this note.
| Z_1x1, 0.5), | ||
| GaussianConditional::sharedMeanAndStddev(Z(i), I_1x1, X(0), | ||
| Z_1x1, 3)}); | ||
| HybridGaussianConditional::Conditionals(modes, conditionals)); |
There was a problem hiding this comment.
So, this is the case of a single key, so here you can use a vector, as we discussed. You would then specify a single DiscreteKey, not a set.
There was a problem hiding this comment.
Ah sorry about that, I had already changed to the DecisionTree constructor before we discussed the single key option. Updated.
There was a problem hiding this comment.
FYI this is for HybridGaussianConditional which can have multiple discrete parents. I've kept this for multiple discrete keys mainly for the wrapper.
Should I still go ahead and update this?
There was a problem hiding this comment.
I'm going ahead and doing this in the interest of time. If a use case comes up where we need multiple discrete parents + the vector of conditionals, we can re-evaluate then. :)
| KeyVector{X(1)}, KeyVector{}, DiscreteKeys{Asia}, | ||
| std::vector{conditional0, conditional1}); | ||
| KeyVector{X(1)}, KeyVector{}, discreteParents, | ||
| HybridGaussianConditional::Conditionals( |
| KeyVector{X(1)}, KeyVector{}, DiscreteKeys{Asia}, | ||
| std::vector{conditional0, conditional1}); | ||
| KeyVector{X(1)}, KeyVector{}, discreteParents, | ||
| HybridGaussianConditional::Conditionals( |
| DiscreteKeys discreteKeys{DiscreteKey(M(0), 2)}; | ||
| HybridNonlinearFactor::Factors factors( | ||
| discreteKeys, {{zero_motion, 0.0}, {one_motion, 0.0}}); | ||
| nfg.emplace_shared<HybridNonlinearFactor>(KeyVector{X(0), X(1)}, discreteKeys, |
HybridNonlinearFactorandHybridGaussianFactornow take a DecisionTree of factor and scalar pairs. This allows us to specify arbitrary values which will be used within the definition of the factors, allowing for relinearization.