John Sorkin
2015-Jan-01 04:52 UTC
[R] normalmixEM does not produce the same results when run twice using the same data.
Windows 7
I am running normalmixEM to fit two normal curves to my data. For some reason,
that I don't understand, the results of running the function twice, on the
same data, results in two different (although similar) sets of values. I would
expect the function run twice on the same data would give exactly the same
results!
I ran the program below and got the following results:
lambda1 mu1 lambda1 lambda2 lambda2 mu2
0.5030387 31.5172824 0.5030387 0.4969613 0.4969613 104.9249404
> run2
lambda1 mu1 lambda1 lambda2 lambda2 mu2
0.5030406 31.5174149 0.5030406 0.4969594 0.4969594 104.9251556
#Define data
John<-c(9.181 ,9.464 ,9.464 ,10.265 ,11.477 ,11.477 ,11.477 ,12.361,
12.361 ,12.361 ,12.361 ,12.985 ,12.985 ,13.384 ,13.384 ,13.962,
14.517 ,14.698 ,14.698 ,14.698 ,14.698 ,15.398 ,15.398 ,16.231,
16.231 ,16.552 ,16.552 ,16.711 ,16.711 ,17.481 ,17.481 ,17.481,
17.927 ,18.506 ,18.506 ,18.928 ,18.928 ,18.928 ,19.612 ,19.612,
19.745 ,19.745 ,20.530 ,20.530 ,21.162 ,21.162 ,21.162 ,21.162,
21.162 ,21.776 ,21.776 ,22.607 ,22.607 ,22.723 ,22.954 ,23.408,
23.632 ,25.353 ,25.353 ,25.663 ,25.663 ,25.663 ,25.969 ,26.171,
26.171 ,26.171 ,26.768 ,27.544 ,27.640 ,27.640 ,27.735 ,28.761,
29.215 ,29.840 ,30.190 ,30.881 ,30.881 ,31.220 ,31.220 ,31.220,
31.888 ,31.888 ,32.135 ,32.461 ,32.461 ,32.623 ,33.104 ,34.123,
35.855 ,35.928 ,35.928 ,36.293 ,36.293 ,37.012 ,37.366 ,37.366,
37.366 ,37.647 ,37.850 ,37.856 ,37.856 ,38.202 ,38.202 ,38.954,
39.021 ,39.089 ,39.089 ,39.223 ,39.491 ,39.624 ,40.153 ,40.609,
40.868 ,41.380 ,41.380 ,42.200 ,42.324 ,42.324 ,42.324 ,43.126,
43.853 ,43.853 ,44.152 ,44.568 ,45.213 ,45.446 ,45.562 ,46.193,
46.193 ,46.421 ,47.320 ,48.203 ,48.257 ,49.284 ,49.656 ,50.550,
50.550 ,51.938 ,52.342 ,52.342 ,52.342 ,52.393 ,53.586 ,53.733,
54.560 ,55.184 ,55.280 ,56.691 ,57.384 ,57.384 ,57.522 ,59.856,
59.856 ,60.599 ,61.975 ,65.004 ,68.861 ,69.052 ,69.811 ,70.897,
71.267 ,71.489 ,73.056 ,74.485 ,74.732 ,75.189 ,76.439 ,77.126,
77.670 ,77.975 ,78.514 ,80.338 ,80.600 ,83.173 ,83.205 ,83.553,
84.181 ,84.960 ,86.160 ,86.405 ,86.618 ,87.646 ,89.284 ,89.401,
89.989 ,90.456 ,90.892 ,90.892 ,90.950 ,91.412 ,91.729 ,92.072,
92.273 ,92.330 ,92.529 ,92.870 ,93.661 ,94.194 ,94.640 ,94.807,
95.029 ,95.112 ,95.885 ,96.132 ,96.869 ,97.438 ,97.600 ,98.567,
99.020 ,99.020 ,99.127 ,100.683 ,100.683 ,101.231 ,102.266 ,102.498,
102.883 ,103.113 ,103.291 ,103.800 ,103.927 ,104.332 ,104.685 ,104.685,
104.710 ,105.911 ,106.333 ,107.073 ,107.123 ,107.172 ,107.172 ,107.907,
107.907 ,108.272 ,108.272 ,108.345 ,108.564 ,108.758 ,109.338 ,110.201,
110.201 ,110.559 ,111.130 ,111.438 ,111.556 ,112.004 ,112.004 ,112.004,
112.684 ,112.940 ,112.940 ,113.034 ,113.661 ,113.661 ,113.962 ,113.985,
114.031 ,114.607 ,114.607 ,114.768 ,115.249 ,115.797 ,116.069 ,116.160,
117.041 ,117.154 ,117.983 ,118.473 ,118.473 ,118.495 ,118.495 ,118.762,
119.027 ,120.501 ,120.959 ,122.000 ,122.022 ,123.396 ,123.609 ,123.737,
124.035 ,124.416 ,124.818 ,124.881 ,124.881 ,125.071 ,125.387 ,125.576,
127.636 ,128.089 ,128.233 ,132.675 ,133.566 ,134.509 ,134.666 ,135.037,
135.504 ,136.589 ,138.675 ,141.961 ,144.608 ,155.222 ,158.646 ,168.502)
# First run of normalmixEM
mixmdl = normalmixEM(John)
run1<-c(lambda1=mixmdl$lambda[1],mu1=mixmdl$mu[1],lambda=mixmdl$lambda,lambda2=mixmdl$lambda[2],mu2=mixmdl$mu[2])
#Second run of normalmixEM
mixmdl = normalmixEM(John)
# Results are not the same!
run2<-c(lambda1=mixmdl$lambda[1],mu1=mixmdl$mu[1],lambda=mixmdl$lambda,lambda2=mixmdl$lambda[2],mu2=mixmdl$mu[2])
run1
run2
John David Sorkin M.D., Ph.D.
Professor of Medicine
Chief, Biostatistics and Informatics
University of Maryland School of Medicine Division of Gerontology and Geriatric
Medicine
Baltimore VA Medical Center
10 North Greene Street
GRECC (BT/18/GR)
Baltimore, MD 21201-1524
(Phone) 410-605-7119
(Fax) 410-605-7913 (Please call phone number above prior to faxing)
Confidentiality Statement:
This email message, including any attachments, is for the sole use of the
intended recipient(s) and may contain confidential and privileged information.
Any unauthorized use, disclosure or distribution is prohibited. If you are not
the intended recipient, please contact the sender by reply email and destroy all
copies of the original message.
Berend Hasselman
2015-Jan-01 10:01 UTC
[R] normalmixEM does not produce the same results when run twice using the same data.
> On 01-01-2015, at 05:52, John Sorkin <JSorkin at grecc.umaryland.edu> wrote: > > Windows 7 > > > I am running normalmixEM to fit two normal curves to my data. For some reason, that I don't understand, the results of running the function twice, on the same data, results in two different (although similar) sets of values. I would expect the function run twice on the same data would give exactly the same results! > > > I ran the program below and got the following results: > > > lambda1 mu1 lambda1 lambda2 lambda2 mu2 > 0.5030387 31.5172824 0.5030387 0.4969613 0.4969613 104.9249404 >> run2 > lambda1 mu1 lambda1 lambda2 lambda2 mu2 > 0.5030406 31.5174149 0.5030406 0.4969594 0.4969594 104.9251556 > > > > > > #Define data > John<-c(9.181 ,9.464 ,9.464 ,10.265 ,11.477 ,11.477 ,11.477 ,12.361, > 12.361 ,12.361 ,12.361 ,12.985 ,12.985 ,13.384 ,13.384 ,13.962, > 14.517 ,14.698 ,14.698 ,14.698 ,14.698 ,15.398 ,15.398 ,16.231, > 16.231 ,16.552 ,16.552 ,16.711 ,16.711 ,17.481 ,17.481 ,17.481, > 17.927 ,18.506 ,18.506 ,18.928 ,18.928 ,18.928 ,19.612 ,19.612, > 19.745 ,19.745 ,20.530 ,20.530 ,21.162 ,21.162 ,21.162 ,21.162, > 21.162 ,21.776 ,21.776 ,22.607 ,22.607 ,22.723 ,22.954 ,23.408, > 23.632 ,25.353 ,25.353 ,25.663 ,25.663 ,25.663 ,25.969 ,26.171, > 26.171 ,26.171 ,26.768 ,27.544 ,27.640 ,27.640 ,27.735 ,28.761, > 29.215 ,29.840 ,30.190 ,30.881 ,30.881 ,31.220 ,31.220 ,31.220, > 31.888 ,31.888 ,32.135 ,32.461 ,32.461 ,32.623 ,33.104 ,34.123, > 35.855 ,35.928 ,35.928 ,36.293 ,36.293 ,37.012 ,37.366 ,37.366, > 37.366 ,37.647 ,37.850 ,37.856 ,37.856 ,38.202 ,38.202 ,38.954, > 39.021 ,39.089 ,39.089 ,39.223 ,39.491 ,39.624 ,40.153 ,40.609, > 40.868 ,41.380 ,41.380 ,42.200 ,42.324 ,42.324 ,42.324 ,43.126, > 43.853 ,43.853 ,44.152 ,44.568 ,45.213 ,45.446 ,45.562 ,46.193, > 46.193 ,46.421 ,47.320 ,48.203 ,48.257 ,49.284 ,49.656 ,50.550, > 50.550 ,51.938 ,52.342 ,52.342 ,52.342 ,52.393 ,53.586 ,53.733, > 54.560 ,55.184 ,55.280 ,56.691 ,57.384 ,57.384 ,57.522 ,59.856, > 59.856 ,60.599 ,61.975 ,65.004 ,68.861 ,69.052 ,69.811 ,70.897, > 71.267 ,71.489 ,73.056 ,74.485 ,74.732 ,75.189 ,76.439 ,77.126, > 77.670 ,77.975 ,78.514 ,80.338 ,80.600 ,83.173 ,83.205 ,83.553, > 84.181 ,84.960 ,86.160 ,86.405 ,86.618 ,87.646 ,89.284 ,89.401, > 89.989 ,90.456 ,90.892 ,90.892 ,90.950 ,91.412 ,91.729 ,92.072, > 92.273 ,92.330 ,92.529 ,92.870 ,93.661 ,94.194 ,94.640 ,94.807, > 95.029 ,95.112 ,95.885 ,96.132 ,96.869 ,97.438 ,97.600 ,98.567, > 99.020 ,99.020 ,99.127 ,100.683 ,100.683 ,101.231 ,102.266 ,102.498, > 102.883 ,103.113 ,103.291 ,103.800 ,103.927 ,104.332 ,104.685 ,104.685, > 104.710 ,105.911 ,106.333 ,107.073 ,107.123 ,107.172 ,107.172 ,107.907, > 107.907 ,108.272 ,108.272 ,108.345 ,108.564 ,108.758 ,109.338 ,110.201, > 110.201 ,110.559 ,111.130 ,111.438 ,111.556 ,112.004 ,112.004 ,112.004, > 112.684 ,112.940 ,112.940 ,113.034 ,113.661 ,113.661 ,113.962 ,113.985, > 114.031 ,114.607 ,114.607 ,114.768 ,115.249 ,115.797 ,116.069 ,116.160, > 117.041 ,117.154 ,117.983 ,118.473 ,118.473 ,118.495 ,118.495 ,118.762, > 119.027 ,120.501 ,120.959 ,122.000 ,122.022 ,123.396 ,123.609 ,123.737, > 124.035 ,124.416 ,124.818 ,124.881 ,124.881 ,125.071 ,125.387 ,125.576, > 127.636 ,128.089 ,128.233 ,132.675 ,133.566 ,134.509 ,134.666 ,135.037, > 135.504 ,136.589 ,138.675 ,141.961 ,144.608 ,155.222 ,158.646 ,168.502) > > > # First run of normalmixEM > mixmdl = normalmixEM(John) > run1<-c(lambda1=mixmdl$lambda[1],mu1=mixmdl$mu[1],lambda=mixmdl$lambda,lambda2=mixmdl$lambda[2],mu2=mixmdl$mu[2]) > #Second run of normalmixEM > mixmdl = normalmixEM(John) > > > # Results are not the same! > run2<-c(lambda1=mixmdl$lambda[1],mu1=mixmdl$mu[1],lambda=mixmdl$lambda,lambda2=mixmdl$lambda[2],mu2=mixmdl$mu[2]) > run1 > run2 >Not surprising if you read the help for normalmixEM. Starting values for three parameters (lambda, mu, sigma) are randomly generated. If you insert set.seed(<some number>) before each call of normalmixEM you will get identical results. Berend