search

baudren / montepython_public

Public repository for the Monte Python Code
MIT License
65 stars 117 forks source link

Size chain. #120

Open Karen-Macias opened 6 years ago

Karen-Macias commented 6 years ago

MontePython gives me an error that no decently sized chain is found, I honestly don't have a clue, maybe it means I have to run it with more points? It used to work -N 100 though.

I appreciate all comments, thanks in advanced.

lesgourg commented 6 years ago

MontePython detects the "burn in" at the beginning of each chain (roughly speaking, the phase in which the likelihood is regularly decreasing towards the best fit, instead of fluctuating in a small range above it). Technically the burn in is defined by MP as "all the points until the step at which the effective chi2 becomes for the first time smaller than the minimum chi2 in the chain plus six". If you have 100 points, they could all be identified as burn in, removed, and nothing will remain apart from a couple of points. Then the code returns the error message that you saw. So you must overcome burn in. In addition, note that all the result from MCMC parameter extraction that you see in the literature are based on runs with about 10'000 to 1'000'000 points after burn in... 100 is nothing!

lesgourg commented 6 years ago

I forgot to say: if you are sending an issue to this website it means that you are not aware that the official montepython_public GitHub has moved to https://github.com/brinckmann/montepython_public There you will find a much more up to date version of the code. please submit your next issues there.

Karen-Macias commented 6 years ago

Thank you so much! I'm just not sure if my laptop will survive 10,000+ points, I will try though. Thanks again.

El lun., 22 de octubre de 2018 0:33, lesgourg notifications@github.com escribió:

MontePython detects the "burn in" at the beginning of each chain (roughly speaking, the phase in which the likelihood is regularly decreasing towards the best fit, instead of fluctuating in a small range above it). Technically the burn in is defined by MP as "all the points until the step at which the effective chi2 becomes for the first time smaller than the minimum chi2 in the chain plus six". If you have 100 points, they could all be identified as burn in, removed, and nothing will remain apart from a couple of points. Then the code returns the error message that you saw. So you must overcome burn in. In addition, note that all the result from MCMC parameter extraction that you see in the literature are based on runs with about 10'000 to 1'000'000 points after burn in... 100 is nothing!

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/baudren/montepython_public/issues/120#issuecomment-431759470, or mute the thread https://github.com/notifications/unsubscribe-auth/Alznh08X4U6LPqM-sFLdvOdfzcAWJn7Eks5unXTmgaJpZM4Xy1Tf .