Somehow the particle filtering tests started failing on Julia 1.6 after merging #527 (job log):
default proposal: Test Failed at /home/runner/work/Gen.jl/Gen.jl/test/inference/particle_filter.jl:169
Expression: isapprox(expected_log_ml, actual_log_ml_est, atol = 0.02)
Evaluated: isapprox(-4.876450833517039, -4.899590122123664; atol = 0.02)
This is strange given that the CI was passing prior to the merge. My guess is that something affected the initial RNG state after the merge, and that we can fix this by increasing the number of particles (to ensure the log ML estimate is more stable).
Somehow the particle filtering tests started failing on Julia 1.6 after merging #527 (job log):
This is strange given that the CI was passing prior to the merge. My guess is that something affected the initial RNG state after the merge, and that we can fix this by increasing the number of particles (to ensure the log ML estimate is more stable).
@limarta do you think you could look into this?