We were segfaulting the DynamicsFitter on very large batches of trials, because trials that weren't being solved as part of the bilevel problem (because we had to sub-sample trials) were getting no joint center information, and that was leading to crashes. This should (in theory at least) fix that problem.
We were segfaulting the DynamicsFitter on very large batches of trials, because trials that weren't being solved as part of the bilevel problem (because we had to sub-sample trials) were getting no joint center information, and that was leading to crashes. This should (in theory at least) fix that problem.