New docs: hybrid #2120
New docs: hybrid #2120
Conversation
|
@varunagrawal FYI |
There was a problem hiding this comment.
Okay this PR is a very nice effort, but I am not comfortable with the wrapping for 2 reasons:
- Lack of tests: whatever I wrapped, I extensively unit tested, and my own work allowed for integration testing.
- Verification with the Matlab wrapper. The matlab wrapper is more finicky than the python wrapper since it cares a whole lot more about return types and other smaller details. I don't see if this was checked (CI doesn't do it).
Moreover, I want to land other hybrid related PRs first (such as DCSAM) and include a new class called Key[Algebraic]DecisionTree which specializes AlgebraicDecisionTree<Key> to make its use much easier and cleaner. It's a low priority item since it simplifies the API but is a lot of effort for a single developer (aka me).
I am not selecting Request changes (which would be the right way to go), but I DO NOT recommend merging this yet.
| uses: pierotofy/set-swap-space@master | ||
| with: | ||
| swap-size-gb: 8 | ||
| shell: bash |
There was a problem hiding this comment.
Can you please document what's going on here? It's much more complicated than what we had before and is not quite obvious.
|
|
||
| /* ************************************************************************ */ | ||
| double HybridConditional::error(const HybridValues &values) const { | ||
| double HybridConditional::error(const HybridValues &hybridValues) const { |
There was a problem hiding this comment.
This change is a bit superficial since HybridConditional::error only ever accepts HybridValues.
| template <> | ||
| struct traits<HybridFactor> : public Testable<HybridFactor> {}; | ||
|
|
||
| // For wrapper: |
There was a problem hiding this comment.
The reason I haven't wrapped a lot of stuff in hybrid is primarily because of this class, and also because I want to add a new class called KeyDecisionTree which specializes the template AlgebraicDecisionTree.
This would have many benefits since we wouldn't have to always pass in keyFormatter arguments, and wrapping would be a cinch, but it is a big PR which is lower priority than some others (such as DCSAM).
| DiscreteKeys newFactorDiscreteKeys; // For the new, restricted factor | ||
|
|
||
| // Iterate over the discrete keys of the current factor | ||
| for (const DiscreteKey& factor_dk_pair : currentFactorDiscreteKeys) { |
There was a problem hiding this comment.
Nitpick: The iterate can simply be called discreteKey.
| const Key& key = factor_dk_pair.first; | ||
|
|
||
| // Check if this key is specified in the assignment | ||
| auto assignment_it = assignment.find(key); |
There was a problem hiding this comment.
This can be combined with the below line
if (assignment.find(key) != assignment.end()) {and since assignment is a std::map, you can just do assignment.at(key) to get the assigned index. You shouldn't need the iterator assignment_it.
| const std::vector<std::pair<gtsam::Vector, double>>& parameters); | ||
|
|
||
| // Standard API | ||
| gtsam::GaussianConditional::shared_ptr choose( |
There was a problem hiding this comment.
shared_ptr has specific wrapper syntax: gtsam::GaussianConditional*.
Also did you test the matlab wrapper?
| const gtsam::VectorValues& frontals) const; | ||
| double logProbability(const gtsam::HybridValues& values) const; | ||
| double evaluate(const gtsam::HybridValues& values) const; | ||
| // double operator()(const gtsam::HybridValues &values) const; |
There was a problem hiding this comment.
This should just work right?
| } | ||
|
|
||
| /* ************************************************************************* */ | ||
| TEST(HybridBayesTree, EliminateMultifrontal) { |
There was a problem hiding this comment.
This test doesn't seem to be verifying correctness. Seems more like a test for dot printing.
|
|
||
| // Check discrete keys now empty | ||
| DiscreteKeys expected_dk1; | ||
| GTSAM_PRINT(restricted->discreteKeys()); |
| /** private helper method for saving the Tree to a text file in GraphViz format */ | ||
| void dot(std::ostream &s, sharedClique clique, const KeyFormatter& keyFormatter, | ||
| int parentnum = 0) const; | ||
| size_t parentnum = 0) const; |
There was a problem hiding this comment.
This may be problematic on different compilers due to the platform dependent size_t definitions.
3 participants
New docs for hybrid folder. There were a lot of classes in here that had many, many functions that were not yet wrapped. Frank also exposed
AlgebraicDecisionTreeKeyfor methods like errorTree().Finished Notebooks