Closed jchodera closed 9 years ago
The automatic LaTeX table generation is pretty awesome, but I will need to generate a special table to replace the table in the paper for the synthetic three-state example. I've renamed the command-line executable script as synthetic-example.py
and added a README.md
to this example. generate-figure.py
will be modified to generate the figure and table data needed for the paper.
ok
Am 17/05/15 um 22:14 schrieb John Chodera:
The automatic LaTeX table generation is pretty awesome, but I will need to generate a special table to replace the table in the paper for the synthetic three-state example. I've renamed the command-line executable script as |synthetic-example.py| and added a |README.md| to this example. |generate-figure.py| will be modified to generate the figure and table data needed for the paper.
— Reply to this email directly or view it on GitHub https://github.com/bhmm/bhmm/pull/36#issuecomment-102846527.
Prof. Dr. Frank Noe Head of Computational Molecular Biology group Freie Universitaet Berlin
Phone: (+49) (0)30 838 75354 Web: research.franknoe.de
I've revised the synthetic three-state model, but both the MLHMM and 1000 cycles of BHMM sampling seems to converge to a different model than the true model. Could you take a look at Table I in the updated manuscript here, @franknoe? Is my revised examples/synthetic-three-state-model/generate-figure.py
script using the BHMM sampling or analysis incorrectly?
I'll check later today, but my day is very full, so it will take a while
Am 18/05/15 um 06:30 schrieb John Chodera:
I've revised the synthetic three-state model, but both the MLHMM and 1000 cycles of BHMM sampling seems to converge to a different model than the true model. Could you take a look at Table I in the updated manuscript here, @franknoe https://github.com/franknoe? Is my revised |examples/synthetic-three-state-model/generate-figure.py| script using the BHMM sampling or analysis incorrectly?
— Reply to this email directly or view it on GitHub https://github.com/bhmm/bhmm/pull/36#issuecomment-102919682.
Prof. Dr. Frank Noe Head of Computational Molecular Biology group Freie Universitaet Berlin
Phone: (+49) (0)30 838 75354 Web: research.franknoe.de
Looks good to me. Below find the true model, the MLE and the BHMM results after 100 steps. As far as I can see all quantities except are within 2-sigma uncertainties. Perhaps you got confused because the order of states has changed? I believe this happens in the Gaussian mixture model estimation.
nstates: 3 Tij: [[ 0.989 0.01 0.001] [ 0.01 0.94 0.05 ] [ 0.001 0.05 0.949]] Pi: [ 0.33333333 0.33333333 0.33333333]
GaussianOutputModel nstates: 3 means: [ 3. 4.7 5.6]
maximum likelihood HMM:
nstates: 3 Tij: [[ 0.98939172 0.0010516 0.00955668] [ 0.00106617 0.9473462 0.05158762] [ 0.00945983 0.05036699 0.94017318]] Pi: [ 0.33371472 0.32915414 0.33713114]
GaussianOutputModel nstates: 3 means: [ 3.00233317 5.59881947 4.7001064 ]
BHMM: [NOW STATE ORDER IS 0,2,1]
Am 18/05/15 um 06:30 schrieb John Chodera:
I've revised the synthetic three-state model, but both the MLHMM and 1000 cycles of BHMM sampling seems to converge to a different model than the true model. Could you take a look at Table I in the updated manuscript here, @franknoe https://github.com/franknoe? Is my revised |examples/synthetic-three-state-model/generate-figure.py| script using the BHMM sampling or analysis incorrectly?
— Reply to this email directly or view it on GitHub https://github.com/bhmm/bhmm/pull/36#issuecomment-102919682.
Prof. Dr. Frank Noe Head of Computational Molecular Biology group Freie Universitaet Berlin
Phone: (+49) (0)30 838 75354 Web: research.franknoe.de
Merging.
Working on updating the manuscript with latest examples.