All 0 output in infer and question about L_thresh

[zhuyixin43]

Hi, I noticed that when I am running -infer I always get all zero parameters in the final_marginals.txt. This leads to no difference in the generated sequence for all my samples. The documentation suggested me to lower my --L_thresh. If I'm not mistaken, --L_thresh is set to default 1e-60 right? I tried to change it to different values such as 1e-50 or 1e-70, but the problem persisted and some samples' running processes gets killed. I also check the "mean number of errors" in the inference_logs.txt, which is all zeros, so I don't think it's the alignment issue? Could you give me some guidance on how to select a reasonable --L_thresh value? Should I be looking at the inference_logs.txt "seq_best_senario" value and use it to get a general idea of my sequence's probability?

Thanks a lot!