[R] FW: help with mclust

Lo, Ken Ken.Lo at roswellpark.org
Thu Oct 4 00:47:27 CEST 2007


> No HTML this time.  Sorry

Dear all,

I am attempting to model some one-dimensional data using Gaussian mixture model with mclust.  Generally, the data that I have have 3 overlapping populations (with one of them being the majority, and the other two combining to less than 15%) and for some reason, mclust consistently ignores the smaller peaks, giving me strange values for the means (completely anti-intuitive in terms of visual inspection by using plot(density())).  However, when I use rnorm to generate some fake data, mclust works consistently well.  As I am quite new to R, I was wondering whether anyone could help me out.  I am using the em function in mclust, with parameters that are arbitrarily set 

Pro <- c(0.33,0.33,0.33)
Mean<- c(-0.5,0,0.5)
Sigmasq <- c(0.05,0.05,0.05)

And a sample of my data set is

c(-0.016133,-0.07668,-0.000625,0.031329,-0.011094,0.014199,-0.014141,0.036836,-0.007695,-0.011738,-0.021953,0.046565,0.010859,-0.050313,-0.167813,0.01543,-0.057598,-0.034336,0.182275,0.032959,0.01918,0.009248,-0.195273,-0.00918,-0.017813,0.003828,-0.113867,0.004014,0.031504,0.004277,0.052188,-0.030859,-0.214023,-0.329102,-0.07832,-0.008379,0.05833,-0.007285,-0.036992,0.035768,0.055006,-0.000781,0.005067,-0.025811,0.021016,-0.002598,-0.036799,-0.03119,-0.004482,-0.024473,-0.108115,-0.116318,0.158008,0.04252,-0.032129,0.00707,-0.073398,-0.115605,-0.033945,-0.022793,0.041855,0.006289,0.250273,0.042607,-0.000449,0.030098,0.041238,-0.028926,-0.111895,0.003867,0.015625,-0.018906,-0.00291,-0.027188,0.00957,-0.133369,0.018652,0.138652,0.038789,-0.050107,0.135908,-0.052725,0.005977,0.030977,0.005371,-0.179902,-0.008691,0.033711,0.164033,-0.063457,0.022734,-0.047227,0.025918,-0.005557,-0.104453,0.021348,-0.054902,-0.069277,-0.115273,0.038906,0.171211,0.000645,-0.064873,0.014062,-0.00252,-0.017715,-0.000586,0.174609,-0.056396,0.000937,-0.217148,-0.203105,0.006533,0.015371,-0.024629,0.015244,0.002949,0.024805,-0.104023,0.007964,-0.198633,-0.007833,-0.000518,0.018232,0.000195,0.028575,-0.028145,-0.030098,-0.002148,-0.035723,-0.005996,0.023027,0.034512,0.009189,0.049252,-0.016641,-0.023262,-0.013379,-0.013633,0.150996,0.040391,0.153809,0.001182,0.040371,-0.016191,-0.05097,-0.12041,0.042617,0.018828,-0.002617,-0.043887,-0.025764,-0.016836,0.023535,0.040625,0.158789,-0.026934,0.02791,-0.108027,0.037979,-0.011865,0.06127,0.04416,-0.07833,0.019922,0.000685,-0.071885,0.000479,0.006211,-0.030879,-0.009188,-0.061895,-0.069102,0.032051,-0.082637,-0.246484,-0.015586,-0.008555,0.265664,0.050781,0.008242,0.169785,-0.025977,0.017871,-0.239492,0.005234,0.006865,0.007344,0.237861,-0.110742,0.208008,0.189336,0.205469,-0.111729,-0.023438,0.49,-0.028281,0.177988,0.00998,-0.002402,0.161924,0.220859,0.026455,0.019629,-0.015098,-0.110771,-0.014414,0.121211,0.028439,-0.026143,0.024989,-0.060801,-0.023124,-0.012734,0.168398,0.039955,-0.038984,-0.028543,0.157412,-0.015547,-0.012617,0.031607,-0.053437,0.027246,0.003906,-0.218613,0.024902,0.020273,0.011914,0.162051,0.00582,-0.019189,-0.009029,0.001875,0.015273,0.175303,-0.092441,-0.086738,-0.022871,0.027852,-0.108809,0.005938,-0.016543,-0.019288,0.210566,-0.022813,-0.001748,-0.108574,0.164971,-0.075186)

If anyone could help me out, I would be extremely grateful. 

Best,

Ken Lo


**********************************************
Ken Lo, PhD
Post-Doctorate Research Affliliate
Department of Cancer Genetics
Buffalo Life Science Complex, L2-104
Roswell Park Cancer Institute
Elm and Carlton Streets
Buffalo, New York 14263
Telephone:  716-845-3941
Fax:         716-845-3940
E-mail:      Ken.Lo at RoswellPark.org
Web:        www.RoswellPark.org 
 
Located in the
Buffalo Life Science Complex
on the Buffalo Niagara Medical Campus
**********************************************





This email message may contain legally privileged and/or confidential information.  If you are not the intended recipient(s), or the employee or agent responsible for the delivery of this message to the intended recipient(s), you are hereby notified that any disclosure, copying, distribution, or use of this email message is prohibited.  If you have received this message in error, please notify the sender immediately by e-mail and delete this email message from your computer. Thank you.



More information about the R-help mailing list