Cannot to resume from a saved state

60 views

Skip to first unread message

Pavel Rinkman

unread,

Jul 24, 2022, 2:40:04 PM7/24/22

to beast-users

Hello, everyone!

I am trying to resume my analyses because they got consistent during 5E7 links but still did not traverse the 200 ESS border. Usually, I do not have problems with loading checkpoints but I have gotten the next error this time

COMPARING LIKELIHOODS: -12297.123855702515 vs. -7514.222179224823
-12297.123855702515    -7514.222179224823
-7514.222179224823    -12297.123855702515
Exception in thread "Thread-0" java.lang.RuntimeException: Saved lnL does not match recomputed value for loaded state: stored lnL: -7514.222179224823, recomputed lnL: -12297.123855702515 (difference 4782.9016764776925).
Your XML may require the construction of a randomly generated starting tree. Try resuming the analysis by using the same starting seed as for the original BEAST run.
        at dr.app.checkpoint.BeastCheckpointer.checkLoadState(Unknown Source)
        at dr.inference.mcmc.MCMC.chain(Unknown Source)
        at dr.inference.mcmc.MCMC.run(Unknown Source)
        at java.base/java.lang.Thread.run(Thread.java:829)

I cannot understand what is wrong with my state file. I have not edited xml-file so this cannot change likelihood computations.

I use BEAST 1.10.4. There are my model, state, & stdout by this link.

Please, could you help me if you have encountered such problem?
I would be very appreciated!

Best regards,

Pavel Rinkman

unread,

Jul 25, 2022, 3:25:23 AM7/25/22

to beast-users

The problem was resolved by specifying the seed from previous analysis.

Honestly, this is very strange for my comprehension. Seed may determine the topology & parameters of a random new-run tree, but how does it affect to the likelihood calculation of a saved state where the last tree information is stored?