Add print method to Graph and HashGraph #70

efernandez · 2019-06-25T08:13:30Z

This allows to print the HashGraph, which is useful for debugging purposes.

efernandez · 2019-06-25T08:17:14Z

fuse_graphs/src/hash_graph.cpp

+         << "  constraints:\n";
+  for (const auto& constraint : constraints_)
+  {
+    stream << "   - " << *constraint.second << "\n";


Is there any way to pass some extra indentation to *constraint.second? (same for *variable.second below)

The main motivation to have proper indentation is that it looks like we could get YAML compliant outputs that could be use to visualize the graph with graphviz if we create a YAML to dot translator. I wonder if that was your plan?

Is there any way to pass some extra indentation to *constraint.second? (same for *variable.second below)

Not currently. We could modify the print() function to accept an indentation string. The problem with that being that there is no way to set that parameter when using the stream operator. We could drop the stream operator in favor of a toString() method. Then you could:

std::cout << constraint.toString(" ") << std::endl;

we could get YAML compliant outputs

That was not something I had thought of. If the output looks like YAML, it is pure coincidence. Returning valid YAML is interesting. The one downside I see is formatting the matrices. I was planning on updating the covariance output to use Eigen's IOFormatter, to get pretty looking matrices. But it may be possible to create pretty valid YAML matrices and we both win.

Regarding toString, although it sounds like the only option, I'm afraid that returning an std::string would be slower than using streams, which IIRC would manage things efficiently in terms of memory and IO.

Re. Eigen matrix, I think there should be an IOFormatter that should be YAML compliant. I could give that idea a try on another PR.

efernandez · 2019-06-25T08:19:57Z

fuse_graphs/src/hash_graph.cpp

+  for (const auto& variable : variables_)
+  {
+    const auto is_on_hold = variables_on_hold_.find(variable.first) != variables_on_hold_.end();
+    stream << "   - " << (is_on_hold ? "[H] " : "") << *variable.second << "\n";


For variables on hold I'm writting [H] , but I wonder if I could instead add another "field" like on_hold: true (or false). That way it'd be complaint with YAML, which seems to be the format you're trying to follow in all your outputs.

The downside though, is that it'd be harder to see if a variables it's on hold because the on_hold field would be at the bottom of the variable output. But for automated tools it'd be perfectly fine.

I added the "on hold" feature because Ceres supported it. However, it did not do what I thought it would do. It acts as though you placed a prior with complete certainty on the "on hold" variables. This results in the variable values being computed as you might expect, but the uncertainty on the "on hold" variables is zero. As such, I'm not sure what the utility of holding a variable is. So, I wouldn't worry overly about the "on hold" being hard to see. I'm not sure how often it will ever get used.

Thanks for the explanation. I've just changed this to the YAML compliant way: on_hold: true/false.

efernandez · 2019-06-25T08:22:35Z

fuse_graphs/src/hash_graph.cpp

+{
+  stream << "HashGraph\n"
+         << "  constraints:\n";
+  for (const auto& constraint : constraints_)


For now I simply print the constraints and variables, but I'm open to suggestions regarding what we print here. There's another map that has variables and constraints pairs that could be printed as well, either replacing what it's printed now, or maybe printing all (constraints list + variables list + variables to constraints list mapping).

The variables and constraints are what I typically print as well. If anything, I'm wondering if this implementation should get moved to the Graph base class. We can use getConstraints() and getVariables() to access the collections in a generic way.

I think that could be the generic implementation in the base class, but I think we should still allow the user to override the default/base implementation, which I think even allows extending it (by calling the base class implementation explicitly from the child class implementation).

svwilliams

This works for me as-is. If we want to move to a formatted toString() in the future, we can do that in some future PR.

svwilliams · 2019-06-25T12:40:08Z

fuse_graphs/src/hash_graph.cpp

+         << "  constraints:\n";
+  for (const auto& constraint : constraints_)
+  {
+    stream << "   - " << *constraint.second << "\n";


Is there any way to pass some extra indentation to *constraint.second? (same for *variable.second below)

Not currently. We could modify the print() function to accept an indentation string. The problem with that being that there is no way to set that parameter when using the stream operator. We could drop the stream operator in favor of a toString() method. Then you could:

std::cout << constraint.toString(" ") << std::endl;

we could get YAML compliant outputs

That was not something I had thought of. If the output looks like YAML, it is pure coincidence. Returning valid YAML is interesting. The one downside I see is formatting the matrices. I was planning on updating the covariance output to use Eigen's IOFormatter, to get pretty looking matrices. But it may be possible to create pretty valid YAML matrices and we both win.

svwilliams · 2019-06-25T12:44:20Z

fuse_graphs/src/hash_graph.cpp

+  for (const auto& variable : variables_)
+  {
+    const auto is_on_hold = variables_on_hold_.find(variable.first) != variables_on_hold_.end();
+    stream << "   - " << (is_on_hold ? "[H] " : "") << *variable.second << "\n";


I added the "on hold" feature because Ceres supported it. However, it did not do what I thought it would do. It acts as though you placed a prior with complete certainty on the "on hold" variables. This results in the variable values being computed as you might expect, but the uncertainty on the "on hold" variables is zero. As such, I'm not sure what the utility of holding a variable is. So, I wouldn't worry overly about the "on hold" being hard to see. I'm not sure how often it will ever get used.

svwilliams · 2019-06-25T12:47:09Z

fuse_graphs/src/hash_graph.cpp

+{
+  stream << "HashGraph\n"
+         << "  constraints:\n";
+  for (const auto& constraint : constraints_)


The variables and constraints are what I typically print as well. If anything, I'm wondering if this implementation should get moved to the Graph base class. We can use getConstraints() and getVariables() to access the collections in a generic way.

Instead of "[H] ", we do "on_hold: true/false".

efernandez · 2019-06-28T07:33:05Z

@svwilliams Are you waiting for any action from my side to merge this? 😃

ayrton04 · 2019-06-28T08:00:33Z

He's waiting on my review. All of our PRs require two reviews. I'll try to get to it today.

Add print method to Graph and HashGraph

ddb3a79

efernandez commented Jun 25, 2019

View reviewed changes

svwilliams approved these changes Jun 25, 2019

View reviewed changes

Print if variable is on hold in YAML complaint way

d72a853

Instead of "[H] ", we do "on_hold: true/false".

svwilliams requested a review from ayrton04 June 26, 2019 16:41

ayrton04 approved these changes Jun 28, 2019

View reviewed changes

matthewbwhitaker approved these changes Jun 28, 2019

View reviewed changes

svwilliams merged commit fa0bdea into locusrobotics:devel Jun 28, 2019

efernandez deleted the add_print_method_for_graph branch March 11, 2020 10:04

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add print method to Graph and HashGraph #70

Add print method to Graph and HashGraph #70

efernandez commented Jun 25, 2019

efernandez Jun 25, 2019

svwilliams Jun 25, 2019

efernandez Jun 25, 2019

efernandez Jun 25, 2019

svwilliams Jun 25, 2019

efernandez Jun 25, 2019

efernandez Jun 25, 2019

svwilliams Jun 25, 2019

efernandez Jun 25, 2019

svwilliams left a comment

svwilliams Jun 25, 2019

svwilliams Jun 25, 2019

svwilliams Jun 25, 2019

efernandez commented Jun 28, 2019

ayrton04 commented Jun 28, 2019

Add print method to Graph and HashGraph #70

Add print method to Graph and HashGraph #70

Conversation

efernandez commented Jun 25, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

svwilliams left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

efernandez commented Jun 28, 2019

ayrton04 commented Jun 28, 2019