Hi community, I am trying to model 5 dependent variables (Observed mean response time comp, Observed mean response time incomp, Predicted mean decision time comp, Predicted mean decision time incomp, Predicted mean nondecision time) using an exponential model with the decay rate (beta) constrained …

Likely after the exponential transformation some values becomes too large, and the gradient at the tail is usually near 0. You should: check the test_value to see if they are indeed too large, if so set more informative prior, and specify test value so they dont blow up initially.

Many thanks for the help. I called print Model_1.test_point just before calling pm.sample and got the following output (the chain crashed again) : {'α_meanRT_c_log__': array(0.69315), 'α_meanRT_i_log__': array(0.69315), 'α_meanDT_c_log__': array(0.69315), 'α_meanDT_i_log__': array(0.69315), 'α_mean…

Sure, below are the results of the debugging: make sure all test_value are finite Model_1.test_point {'α_meanRT_c_log__': array(0.69315), 'α_meanRT_i_log__': array(0.69315), 'α_meanDT_c_log__': array(0.69315), 'α_meanDT_i_log__': array(0.69315), 'α_meanTer_log__': array(0.69315), 'β': array(0.334),…

Glad you solved it. Yes the jitter sometimes could cause problem, especially if you have random variable in a small scale, jitter might make the proposal/test value invalid.

Ok Thanks. I just tried to compare the above model with a similar model that allows the decay parameter of the exponential function free to vary across my dependent variables. I got some warnings during LOO and WAIC computations: /gpfs22/home/servanm/myenv/lib/python3.6/site-packages/pymc3/stats.py…

That’s not directly related to the previous error, but it is something to do with prior in some way. It is a bit more difficult to get rid of because you would need to play with the scale of your prior to see what works.

Chain failure with an exponential model

Questions

junpenglao September 17, 2019, 3:22pm 4

Could you try the debugging step here: How to track a 'nan energy'?

Topic		Replies	Views
Chain failure causing PyMC3 to error out Questions	1	5167	September 10, 2018
Fitting a complex model with a large dataset: chain contains only diverging samples Questions	5	1339	December 12, 2019
Problem with log link in fractional polynomial model	10	313	April 29, 2024
Chain Failure when running model Questions	1	1956	March 7, 2020
Multiplicative Factor Model (degradations) version agnostic	10	409	February 21, 2024

Chain failure with an exponential model

Related topics