The problem: If the user's solution to a puzzle is a rank-0 tensor (eg, tensor(6)), running make_test results in a cryptic TypeError many levels deep in the SVG-rendering code. This is documented pretty well by @ikamensh in #23.
This solution: This change adds a check, torch.is_tensor(yours) and yours.ndim > 0, before adding the user's solution output to the data to be rendered to the SVG.
Caveats: The solution checker seems to accept tensor(6), tensor([6]), tensor([[6]]), equally. Before this change, the SVG-rendering code would error out before solution checker ran, so you could call that "failing". I don't completely understand the solution-checking code and I'm not sure whether that is intended behavior.
Hopefully this makes sense. I'm happy to discuss it or make changes if you'd like. Thanks in advance!
The problem: If the user's solution to a puzzle is a rank-0 tensor (eg,
tensor(6)), runningmake_testresults in a cryptic TypeError many levels deep in the SVG-rendering code. This is documented pretty well by @ikamensh in #23.This solution: This change adds a check,
torch.is_tensor(yours) and yours.ndim > 0, before adding the user's solution output to the data to be rendered to the SVG.Caveats: The solution checker seems to accept
tensor(6),tensor([6]),tensor([[6]]), equally. Before this change, the SVG-rendering code would error out before solution checker ran, so you could call that "failing". I don't completely understand the solution-checking code and I'm not sure whether that is intended behavior.Hopefully this makes sense. I'm happy to discuss it or make changes if you'd like. Thanks in advance!