The R Project SVN
R-packages
(root)/trunk/Old/rpart/tests/testall.Rout.save – Rev 7500
Rev 6385 | Blame | Compare with Previous | Last modification | View Log | Download | RSS feed
R Under development (unstable) (2012-11-27 r61172) -- "Unsuffered Consequences"Copyright (C) 2012 The R Foundation for Statistical ComputingISBN 3-900051-07-0Platform: x86_64-apple-darwin11.4.2/x86_64 (64-bit)R is free software and comes with ABSOLUTELY NO WARRANTY.You are welcome to redistribute it under certain conditions.Type 'license()' or 'licence()' for distribution details.R is a collaborative project with many contributors.Type 'contributors()' for more information and'citation()' on how to cite R or R packages in publications.Type 'demo()' for some demos, 'help()' for on-line help, or'help.start()' for an HTML browser interface to help.Type 'q()' to quit R.> #> # This is a set of tests, with printout, that tries> # to exercise all of the splitting rules> # It's less formal than the others in this directory, but covers> # more.> # For all of the cross-validations I set xgroups explicitly, so as> # to avoid any changes in random number allocation of the groups.>> library(rpart)> require(survival)Loading required package: survivalLoading required package: splines> options(digits=4) #avoid trivial rounding changes across R versions>> #> #> # Survival, using the stagec data> #> # Time to progression in years> # status 1=progressed, 0= censored> # age> # early endocrine therapy 1=no 2=yes> # % of cells in g2 phase, from flow cytometry> # tumor grade (Farrow) 1,2,3,4> # Gleason score (competing grading system)> # ploidy> xgroup <- rep(1:10, length=nrow(stagec))> fit1 <- rpart(Surv(pgtime, pgstat) ~ age + eet + g2+grade+gleason +ploidy,+ stagec, method="poisson",+ control=rpart.control(usesurrogate=0, cp=0, xval=xgroup))> fit1n= 146node), split, n, deviance, yval* denotes terminal node1) root 146 193.400 0.0584902) grade< 2.5 61 46.000 0.0212704) g2< 11.36 33 9.303 0.0071748) g2>=6.545 26 1.847 0.004462 *9) g2< 6.545 7 5.883 0.027860 *5) g2>=11.36 27 27.880 0.04487010) g2>=14.37 15 12.430 0.030940 *11) g2< 14.37 12 13.690 0.067870 *3) grade>=2.5 85 122.600 0.0944006) age>=56.5 75 104.700 0.08284012) g2>=22.77 12 14.830 0.039530 *13) g2< 22.77 57 76.110 0.10030026) gleason< 7.5 41 55.810 0.07686052) g2< 13.47 23 26.800 0.050830104) g2>=11.52 7 5.289 0.030820 *105) g2< 11.52 16 20.270 0.063870 *53) g2>=13.47 18 25.520 0.112300 *27) gleason>=7.5 16 15.190 0.164400 *7) age< 56.5 10 11.420 0.196900 *> summary(fit1)Call:rpart(formula = Surv(pgtime, pgstat) ~ age + eet + g2 + grade +gleason + ploidy, data = stagec, method = "poisson", control = rpart.control(usesurrogate = 0,cp = 0, xval = xgroup))n= 146CP nsplit rel error xerror xstd1 0.128372 0 1.0000 1.0120 0.063292 0.052298 1 0.8716 0.9220 0.071703 0.045608 3 0.7670 0.9728 0.078464 0.026442 4 0.7214 0.9732 0.080525 0.018049 5 0.6950 0.9725 0.083886 0.009068 6 0.6769 0.9698 0.085327 0.008132 7 0.6679 0.9920 0.087958 0.006412 8 0.6597 0.9853 0.088009 0.000000 9 0.6533 1.0077 0.08952Variable importancegrade g2 gleason ploidy age eet27 27 23 12 10 1Node number 1: 146 observations, complexity param=0.1284events=54, estimated rate=0.05849 , mean deviance=1.325left son=2 (61 obs) right son=3 (85 obs)Primary splits:grade < 2.5 to the left, improve=24.880, (0 missing)gleason < 5.5 to the left, improve=21.630, (3 missing)ploidy splits as LRR, improve=13.020, (0 missing)g2 < 13.2 to the left, improve=11.450, (7 missing)age < 53.5 to the right, improve= 2.484, (0 missing)Surrogate splits:gleason < 5.5 to the left, agree=0.863, adj=0.672, (0 split)ploidy splits as LRR, agree=0.644, adj=0.148, (0 split)g2 < 9.945 to the left, agree=0.630, adj=0.115, (0 split)age < 66.5 to the right, agree=0.589, adj=0.016, (0 split)Node number 2: 61 observations, complexity param=0.04561events=9, estimated rate=0.02127 , mean deviance=0.7542left son=4 (33 obs) right son=5 (27 obs), 1 observation remainsPrimary splits:g2 < 11.36 to the left, improve=8.9650, (1 missing)ploidy splits as LRL, improve=7.4700, (0 missing)age < 68.5 to the right, improve=3.9560, (0 missing)gleason < 5.5 to the left, improve=1.8010, (3 missing)eet < 1.5 to the left, improve=0.5444, (1 missing)Surrogate splits:ploidy splits as LR-, agree=0.917, adj=0.815, (0 split)age < 65.5 to the left, agree=0.617, adj=0.148, (0 split)Node number 3: 85 observations, complexity param=0.0523events=45, estimated rate=0.0944 , mean deviance=1.442left son=6 (75 obs) right son=7 (10 obs)Primary splits:age < 56.5 to the right, improve=7.3190, (0 missing)g2 < 22.76 to the right, improve=6.8420, (6 missing)gleason < 8.5 to the left, improve=4.8530, (0 missing)ploidy splits as LLR, improve=1.7320, (0 missing)eet < 1.5 to the right, improve=0.9628, (1 missing)Node number 4: 33 observations, complexity param=0.008132events=1, estimated rate=0.007174 , mean deviance=0.2819left son=8 (26 obs) right son=9 (7 obs)Primary splits:g2 < 6.545 to the right, improve=3.1310, (0 missing)age < 62.5 to the right, improve=1.3520, (0 missing)eet < 1.5 to the left, improve=0.6605, (0 missing)gleason < 5.5 to the right, improve=0.4889, (2 missing)Surrogate splits:grade < 1.5 to the right, agree=0.848, adj=0.286, (0 split)Node number 5: 27 observations, complexity param=0.009068events=8, estimated rate=0.04487 , mean deviance=1.033left son=10 (15 obs) right son=11 (12 obs)Primary splits:g2 < 14.37 to the right, improve=1.820, (0 missing)gleason < 5.5 to the left, improve=1.698, (1 missing)age < 62.5 to the left, improve=1.136, (0 missing)Surrogate splits:ploidy splits as RL-, agree=0.741, adj=0.417, (0 split)age < 65 to the left, agree=0.630, adj=0.167, (0 split)gleason < 5.5 to the left, agree=0.630, adj=0.167, (0 split)Node number 6: 75 observations, complexity param=0.0523events=37, estimated rate=0.08284 , mean deviance=1.397left son=12 (12 obs) right son=13 (57 obs), 6 observations remainPrimary splits:g2 < 22.76 to the right, improve=4.8580, (6 missing)gleason < 7.5 to the left, improve=4.1030, (0 missing)age < 69.5 to the left, improve=2.7250, (0 missing)ploidy splits as LRR, improve=0.7858, (0 missing)eet < 1.5 to the right, improve=0.1992, (1 missing)Node number 7: 10 observationsevents=8, estimated rate=0.1969 , mean deviance=1.142Node number 8: 26 observationsevents=0, estimated rate=0.004462 , mean deviance=0.07105Node number 9: 7 observationsevents=1, estimated rate=0.02786 , mean deviance=0.8404Node number 10: 15 observationsevents=3, estimated rate=0.03094 , mean deviance=0.8289Node number 11: 12 observationsevents=5, estimated rate=0.06787 , mean deviance=1.141Node number 12: 12 observationsevents=4, estimated rate=0.03953 , mean deviance=1.236Node number 13: 57 observations, complexity param=0.02644events=30, estimated rate=0.1003 , mean deviance=1.335left son=26 (41 obs) right son=27 (16 obs)Primary splits:gleason < 7.5 to the left, improve=5.40700, (0 missing)g2 < 13.2 to the left, improve=4.46300, (0 missing)ploidy splits as LRR, improve=3.49400, (0 missing)age < 69.5 to the left, improve=3.12800, (0 missing)eet < 1.5 to the left, improve=0.04903, (1 missing)Surrogate splits:grade < 3.5 to the left, agree=0.772, adj=0.187, (0 split)Node number 26: 41 observations, complexity param=0.01805events=18, estimated rate=0.07686 , mean deviance=1.361left son=52 (23 obs) right son=53 (18 obs)Primary splits:g2 < 13.48 to the left, improve=3.57000, (0 missing)ploidy splits as LRR, improve=3.05800, (0 missing)age < 65.5 to the left, improve=0.68120, (0 missing)eet < 1.5 to the left, improve=0.09244, (0 missing)gleason < 6.5 to the right, improve=0.01761, (0 missing)Surrogate splits:eet < 1.5 to the right, agree=0.634, adj=0.167, (0 split)age < 58.5 to the right, agree=0.585, adj=0.056, (0 split)Node number 27: 16 observationsevents=12, estimated rate=0.1644 , mean deviance=0.9493Node number 52: 23 observations, complexity param=0.006412events=7, estimated rate=0.05083 , mean deviance=1.165left son=104 (7 obs) right son=105 (16 obs)Primary splits:g2 < 11.52 to the right, improve=1.41100, (0 missing)age < 62.5 to the right, improve=1.24600, (0 missing)gleason < 6.5 to the left, improve=0.02047, (0 missing)Surrogate splits:age < 71 to the right, agree=0.826, adj=0.429, (0 split)ploidy splits as RLR, agree=0.739, adj=0.143, (0 split)Node number 53: 18 observationsevents=11, estimated rate=0.1123 , mean deviance=1.418Node number 104: 7 observationsevents=1, estimated rate=0.03082 , mean deviance=0.7555Node number 105: 16 observationsevents=6, estimated rate=0.06387 , mean deviance=1.267>> fit2 <- rpart(Surv(pgtime, pgstat) ~ age + eet + g2+grade+gleason +ploidy,+ stagec, xval=xgroup)> fit2n= 146node), split, n, deviance, yval* denotes terminal node1) root 146 192.100 1.00002) grade< 2.5 61 44.800 0.36344) g2< 11.36 33 9.117 0.1230 *5) g2>=11.36 28 27.600 0.734610) gleason< 5.5 20 14.300 0.5304 *11) gleason>=5.5 8 11.090 1.3070 *3) grade>=2.5 85 122.400 1.61506) age>=56.5 75 103.100 1.426012) gleason< 7.5 50 66.120 1.141024) g2< 13.47 24 27.200 0.8007 *25) g2>=13.47 26 36.790 1.457050) g2>=17.91 15 20.330 0.9790 *51) g2< 17.91 11 13.460 2.1710 *13) gleason>=7.5 25 33.490 2.031026) g2>=15.29 10 11.590 1.2160 *27) g2< 15.29 15 18.940 2.7050 *7) age< 56.5 10 13.770 3.1820 *> summary(fit2)Call:rpart(formula = Surv(pgtime, pgstat) ~ age + eet + g2 + grade +gleason + ploidy, data = stagec, xval = xgroup)n= 146CP nsplit rel error xerror xstd1 0.12946 0 1.0000 1.0100 0.071082 0.04206 1 0.8705 0.8919 0.074203 0.02920 2 0.8285 0.9239 0.080044 0.01799 3 0.7993 0.9293 0.082925 0.01541 4 0.7813 0.9866 0.092146 0.01335 5 0.7659 1.0007 0.091617 0.01151 7 0.7392 0.9983 0.090658 0.01000 8 0.7277 1.0130 0.09176Variable importancegrade gleason g2 ploidy age eet27 25 22 13 11 2Node number 1: 146 observations, complexity param=0.1295events=54, estimated rate=1 , mean deviance=1.316left son=2 (61 obs) right son=3 (85 obs)Primary splits:grade < 2.5 to the left, improve=24.920, (0 missing)gleason < 5.5 to the left, improve=21.350, (3 missing)ploidy splits as LRR, improve=13.790, (0 missing)g2 < 13.2 to the left, improve=12.310, (7 missing)age < 58.5 to the right, improve= 2.684, (0 missing)Surrogate splits:gleason < 5.5 to the left, agree=0.863, adj=0.672, (0 split)ploidy splits as LRR, agree=0.644, adj=0.148, (0 split)g2 < 9.945 to the left, agree=0.630, adj=0.115, (0 split)age < 66.5 to the right, agree=0.589, adj=0.016, (0 split)Node number 2: 61 observations, complexity param=0.04206events=9, estimated rate=0.3634 , mean deviance=0.7344left son=4 (33 obs) right son=5 (28 obs)Primary splits:g2 < 11.36 to the left, improve=8.9620, (1 missing)ploidy splits as LRL, improve=7.5850, (0 missing)age < 68.5 to the right, improve=4.1380, (0 missing)gleason < 5.5 to the left, improve=1.6410, (3 missing)eet < 1.5 to the left, improve=0.5961, (1 missing)Surrogate splits:ploidy splits as LR-, agree=0.917, adj=0.815, (0 split)age < 65.5 to the left, agree=0.617, adj=0.148, (1 split)Node number 3: 85 observations, complexity param=0.0292events=45, estimated rate=1.615 , mean deviance=1.44left son=6 (75 obs) right son=7 (10 obs)Primary splits:age < 56.5 to the right, improve=6.3380, (0 missing)g2 < 22.76 to the right, improve=4.7180, (6 missing)gleason < 8.5 to the left, improve=4.2190, (0 missing)ploidy splits as LRR, improve=1.5590, (0 missing)eet < 1.5 to the right, improve=0.8558, (1 missing)Node number 4: 33 observationsevents=1, estimated rate=0.123 , mean deviance=0.2763Node number 5: 28 observations, complexity param=0.01151events=8, estimated rate=0.7346 , mean deviance=0.9858left son=10 (20 obs) right son=11 (8 obs)Primary splits:gleason < 5.5 to the left, improve=2.090, (1 missing)age < 67.5 to the right, improve=1.639, (0 missing)g2 < 14.37 to the right, improve=1.437, (1 missing)Node number 6: 75 observations, complexity param=0.01799events=37, estimated rate=1.426 , mean deviance=1.374left son=12 (50 obs) right son=13 (25 obs)Primary splits:gleason < 7.5 to the left, improve=3.5210, (0 missing)g2 < 23.48 to the right, improve=3.3340, (6 missing)age < 69.5 to the left, improve=2.2120, (0 missing)ploidy splits as LRR, improve=1.1190, (0 missing)eet < 1.5 to the right, improve=0.1528, (1 missing)Surrogate splits:grade < 3.5 to the left, agree=0.733, adj=0.2, (0 split)Node number 7: 10 observationsevents=8, estimated rate=3.182 , mean deviance=1.377Node number 10: 20 observationsevents=4, estimated rate=0.5304 , mean deviance=0.7149Node number 11: 8 observationsevents=4, estimated rate=1.307 , mean deviance=1.387Node number 12: 50 observations, complexity param=0.01335events=21, estimated rate=1.141 , mean deviance=1.322left son=24 (24 obs) right son=25 (26 obs)Primary splits:g2 < 13.48 to the left, improve=1.87000, (3 missing)ploidy splits as LRR, improve=1.45100, (0 missing)age < 62.5 to the right, improve=0.79680, (0 missing)gleason < 6.5 to the left, improve=0.02209, (0 missing)eet < 1.5 to the right, improve=0.01274, (0 missing)Surrogate splits:age < 66.5 to the right, agree=0.596, adj=0.174, (3 split)eet < 1.5 to the right, agree=0.553, adj=0.087, (0 split)gleason < 6.5 to the left, agree=0.553, adj=0.087, (0 split)Node number 13: 25 observations, complexity param=0.01541events=16, estimated rate=2.031 , mean deviance=1.339left son=26 (10 obs) right son=27 (15 obs)Primary splits:g2 < 15.29 to the right, improve=3.893000, (3 missing)age < 68.5 to the left, improve=1.327000, (0 missing)ploidy splits as RLR, improve=0.858700, (0 missing)eet < 1.5 to the right, improve=0.001483, (1 missing)Surrogate splits:ploidy splits as RLR, agree=0.727, adj=0.4, (3 split)gleason < 8.5 to the right, agree=0.636, adj=0.2, (0 split)age < 57.5 to the left, agree=0.591, adj=0.1, (0 split)eet < 1.5 to the left, agree=0.591, adj=0.1, (0 split)Node number 24: 24 observationsevents=7, estimated rate=0.8007 , mean deviance=1.133Node number 25: 26 observations, complexity param=0.01335events=14, estimated rate=1.457 , mean deviance=1.415left son=50 (15 obs) right son=51 (11 obs)Primary splits:g2 < 17.92 to the right, improve=3.170000, (2 missing)age < 64.5 to the left, improve=0.971200, (0 missing)eet < 1.5 to the right, improve=0.961700, (0 missing)gleason < 6.5 to the left, improve=0.005796, (0 missing)Surrogate splits:age < 65.5 to the left, agree=0.708, adj=0.364, (2 split)eet < 1.5 to the right, agree=0.667, adj=0.273, (0 split)gleason < 6.5 to the right, agree=0.625, adj=0.182, (0 split)Node number 26: 10 observationsevents=5, estimated rate=1.216 , mean deviance=1.159Node number 27: 15 observationsevents=11, estimated rate=2.705 , mean deviance=1.263Node number 50: 15 observationsevents=6, estimated rate=0.979 , mean deviance=1.356Node number 51: 11 observationsevents=8, estimated rate=2.171 , mean deviance=1.224>>> #> # Continuous y variable> # Use deterministic xgroups: the first group is the 1st, 11th, 21st, etc> # smallest obs, the second is 2, 12, 22, ...>> mystate <- data.frame(state.x77, region=factor(state.region))> names(mystate) <- c("population","income" , "illiteracy","life" ,+ "murder", "hs.grad", "frost", "area", "region")>> xvals <- 1:nrow(mystate)> xvals[order(mystate$income)] <- rep(1:10, length=nrow(mystate))>> fit4 <- rpart(income ~ population + region + illiteracy +life + murder ++ hs.grad + frost , mystate,+ control=rpart.control(minsplit=10, xval=xvals))>> summary(fit4)Call:rpart(formula = income ~ population + region + illiteracy + life +murder + hs.grad + frost, data = mystate, control = rpart.control(minsplit = 10,xval = xvals))n= 50CP nsplit rel error xerror xstd1 0.42831 0 1.0000 1.0157 0.22652 0.13514 1 0.5717 0.5948 0.17473 0.06458 2 0.4365 0.7005 0.18024 0.05485 3 0.3720 0.8606 0.22995 0.05346 4 0.3171 0.8494 0.22976 0.02479 5 0.2637 0.8289 0.26447 0.01940 6 0.2389 0.8196 0.26418 0.01394 7 0.2195 0.8196 0.26419 0.01000 8 0.2055 0.8117 0.2640Variable importancehs.grad life illiteracy murder region population frost28 22 13 12 12 9 5Node number 1: 50 observations, complexity param=0.4283mean=4436, MSE=3.7e+05left son=2 (10 obs) right son=3 (40 obs)Primary splits:hs.grad < 44.3 to the left, improve=0.4283, (0 missing)illiteracy < 1.55 to the right, improve=0.3249, (0 missing)region splits as RLRR, improve=0.2285, (0 missing)murder < 11.55 to the right, improve=0.2012, (0 missing)life < 68.9 to the left, improve=0.1672, (0 missing)Surrogate splits:illiteracy < 1.55 to the right, agree=0.90, adj=0.5, (0 split)life < 69.26 to the left, agree=0.90, adj=0.5, (0 split)region splits as RLRR, agree=0.88, adj=0.4, (0 split)murder < 11.55 to the right, agree=0.88, adj=0.4, (0 split)frost < 81 to the left, agree=0.82, adj=0.1, (0 split)Node number 2: 10 observations, complexity param=0.0194mean=3640, MSE=6.688e+04left son=4 (7 obs) right son=5 (3 obs)Primary splits:population < 3990 to the left, improve=0.53670, (0 missing)frost < 55 to the left, improve=0.30250, (0 missing)illiteracy < 2.2 to the right, improve=0.29240, (0 missing)hs.grad < 40.8 to the right, improve=0.14540, (0 missing)murder < 12.85 to the left, improve=0.08289, (0 missing)Node number 3: 40 observations, complexity param=0.1351mean=4635, MSE=2.477e+05left son=6 (34 obs) right son=7 (6 obs)Primary splits:life < 70.26 to the right, improve=0.25230, (0 missing)murder < 10 to the left, improve=0.16790, (0 missing)hs.grad < 60.95 to the left, improve=0.12350, (0 missing)illiteracy < 0.75 to the left, improve=0.11520, (0 missing)population < 659 to the right, improve=0.08106, (0 missing)Surrogate splits:population < 613.5 to the right, agree=0.875, adj=0.167, (0 split)murder < 11.2 to the left, agree=0.875, adj=0.167, (0 split)hs.grad < 64.55 to the left, agree=0.875, adj=0.167, (0 split)Node number 4: 7 observationsmean=3516, MSE=3.844e+04Node number 5: 3 observationsmean=3929, MSE=1.361e+04Node number 6: 34 observations, complexity param=0.06458mean=4530, MSE=1.69e+05left son=12 (12 obs) right son=13 (22 obs)Primary splits:population < 1374 to the left, improve=0.20790, (0 missing)life < 70.41 to the left, improve=0.19010, (0 missing)frost < 33.5 to the right, improve=0.10930, (0 missing)illiteracy < 0.75 to the left, improve=0.08942, (0 missing)hs.grad < 59.7 to the left, improve=0.06429, (0 missing)Surrogate splits:frost < 152 to the right, agree=0.794, adj=0.417, (0 split)life < 70.41 to the left, agree=0.735, adj=0.250, (0 split)murder < 2 to the left, agree=0.706, adj=0.167, (0 split)region splits as RRRL, agree=0.676, adj=0.083, (0 split)illiteracy < 1.85 to the right, agree=0.676, adj=0.083, (0 split)Node number 7: 6 observationsmean=5230, MSE=2.768e+05Node number 12: 12 observations, complexity param=0.05485mean=4276, MSE=1.942e+05left son=24 (3 obs) right son=25 (9 obs)Primary splits:population < 994.5 to the right, improve=0.4355, (0 missing)life < 71.89 to the left, improve=0.2955, (0 missing)frost < 172.5 to the left, improve=0.2333, (0 missing)illiteracy < 0.75 to the left, improve=0.1965, (0 missing)murder < 2.55 to the right, improve=0.1847, (0 missing)Surrogate splits:life < 70.47 to the left, agree=0.833, adj=0.333, (0 split)Node number 13: 22 observations, complexity param=0.05346mean=4668, MSE=1.01e+05left son=26 (4 obs) right son=27 (18 obs)Primary splits:hs.grad < 52.05 to the left, improve=0.44500, (0 missing)region splits as RLLR, improve=0.35050, (0 missing)life < 71.56 to the left, improve=0.12220, (0 missing)murder < 5.65 to the right, improve=0.10520, (0 missing)population < 2980 to the left, improve=0.06276, (0 missing)Surrogate splits:region splits as RLRR, agree=0.864, adj=0.25, (0 split)Node number 24: 3 observationsmean=3772, MSE=3.261e+04Node number 25: 9 observationsmean=4444, MSE=1.353e+05Node number 26: 4 observationsmean=4218, MSE=2.77e+04Node number 27: 18 observations, complexity param=0.02479mean=4768, MSE=6.236e+04left son=54 (8 obs) right son=55 (10 obs)Primary splits:region splits as RRLR, improve=0.4086, (0 missing)illiteracy < 1 to the left, improve=0.3653, (0 missing)murder < 3.05 to the left, improve=0.1805, (0 missing)hs.grad < 52.75 to the right, improve=0.1507, (0 missing)population < 2412 to the left, improve=0.1427, (0 missing)Surrogate splits:illiteracy < 1 to the left, agree=0.833, adj=0.625, (0 split)frost < 108.5 to the right, agree=0.833, adj=0.625, (0 split)life < 72.31 to the right, agree=0.778, adj=0.500, (0 split)murder < 3.05 to the left, agree=0.778, adj=0.500, (0 split)hs.grad < 59.95 to the left, agree=0.667, adj=0.250, (0 split)Node number 54: 8 observationsmean=4590, MSE=1.009e+04Node number 55: 10 observations, complexity param=0.01394mean=4911, MSE=5.832e+04left son=110 (7 obs) right son=111 (3 obs)Primary splits:frost < 109 to the left, improve=0.4423, (0 missing)population < 2820 to the left, improve=0.3546, (0 missing)hs.grad < 57.05 to the right, improve=0.3103, (0 missing)region splits as RL-L, improve=0.2564, (0 missing)illiteracy < 1.2 to the right, improve=0.1921, (0 missing)Surrogate splits:life < 71.94 to the left, agree=0.8, adj=0.333, (0 split)Node number 110: 7 observationsmean=4806, MSE=2.969e+04Node number 111: 3 observationsmean=5156, MSE=3.914e+04>>> #> # Check out xpred.rpart> #> meany <- mean(mystate$income)> xpr <- xpred.rpart(fit4, xval=xvals)> xpr2 <- (xpr - mystate$income)^2> risk0 <- mean((mystate$income - meany)^2)> xpmean <- as.vector(apply(xpr2, 2, mean)) #kill the names> all.equal(xpmean/risk0, as.vector(fit4$cptable[,'xerror']))[1] TRUE>> xpstd <- as.vector(apply((sweep(xpr2, 2, xpmean))^2, 2, sum))> xpstd <- sqrt(xpstd)/(50*risk0)> all.equal(xpstd, as.vector(fit4$cptable[,'xstd']))[1] TRUE>> #> # recreate subset #3 of the xval> #> tfit4 <- rpart(income ~ population + region + illiteracy +life + murder ++ hs.grad + frost , mystate, subset=(xvals!=3),+ control=rpart.control(minsplit=10, xval=0))> tpred <- predict(tfit4, mystate[xvals==3,])> all.equal(tpred, xpr[xvals==3,ncol(xpr)])[1] TRUE>> # How much does this differ from the "real" formula, more complex,> # found on page 309 of Breiman et al. ?> #xtemp <- (xpr2/outer(rep(1,50),xpmean)) - ((mystate$income - meany)^2)/risk0> #real.se<- xpmean* sqrt(apply(xtemp^2,2,sum))/(risk0*50)>>> # Simple yes/no classification model> fit5 <- rpart(factor(pgstat) ~ age + eet + g2+grade+gleason +ploidy,+ stagec, xval=xgroup)>> fit5n= 146node), split, n, loss, yval, (yprob)* denotes terminal node1) root 146 54 0 (0.6301 0.3699)2) grade< 2.5 61 9 0 (0.8525 0.1475) *3) grade>=2.5 85 40 1 (0.4706 0.5294)6) g2< 13.2 40 17 0 (0.5750 0.4250)12) ploidy=diploid,tetraploid 31 11 0 (0.6452 0.3548)24) g2>=11.84 7 1 0 (0.8571 0.1429) *25) g2< 11.84 24 10 0 (0.5833 0.4167)50) g2< 11 17 5 0 (0.7059 0.2941) *51) g2>=11 7 2 1 (0.2857 0.7143) *13) ploidy=aneuploid 9 3 1 (0.3333 0.6667) *7) g2>=13.2 45 17 1 (0.3778 0.6222)14) g2>=17.91 22 8 0 (0.6364 0.3636)28) age>=62.5 15 4 0 (0.7333 0.2667) *29) age< 62.5 7 3 1 (0.4286 0.5714) *15) g2< 17.91 23 3 1 (0.1304 0.8696) *>> fit6 <- rpart(factor(pgstat) ~ age + eet + g2+grade+gleason +ploidy,+ stagec, parm=list(prior=c(.5,.5)), xval=xgroup)> summary(fit6)Call:rpart(formula = factor(pgstat) ~ age + eet + g2 + grade + gleason +ploidy, data = stagec, parms = list(prior = c(0.5, 0.5)),xval = xgroup)n= 146CP nsplit rel error xerror xstd1 0.39855 0 1.0000 1.2081 0.093812 0.02335 1 0.6014 0.6679 0.076843 0.02275 3 0.5548 0.8378 0.084454 0.01892 5 0.5093 0.8857 0.085855 0.01530 7 0.4714 0.8857 0.085856 0.01000 9 0.4408 0.8639 0.08599Variable importanceg2 grade gleason ploidy age eet28 26 20 18 6 2Node number 1: 146 observations, complexity param=0.3986predicted class=0 expected loss=0.5 P(node) =1class counts: 92 54probabilities: 0.500 0.500left son=2 (61 obs) right son=3 (85 obs)Primary splits:grade < 2.5 to the left, improve=12.490, (0 missing)gleason < 5.5 to the left, improve=10.510, (3 missing)ploidy splits as LRR, improve= 9.018, (0 missing)g2 < 13.2 to the left, improve= 8.281, (7 missing)age < 58.5 to the right, improve= 1.520, (0 missing)Surrogate splits:gleason < 5.5 to the left, agree=0.863, adj=0.672, (0 split)ploidy splits as LRR, agree=0.644, adj=0.148, (0 split)g2 < 9.945 to the left, agree=0.630, adj=0.115, (0 split)age < 66.5 to the right, agree=0.589, adj=0.016, (0 split)Node number 2: 61 observations, complexity param=0.0153predicted class=0 expected loss=0.2277 P(node) =0.3659class counts: 52 9probabilities: 0.772 0.228left son=4 (33 obs) right son=5 (28 obs)Primary splits:g2 < 11.36 to the left, improve=3.5470, (1 missing)ploidy splits as LRL, improve=3.2970, (0 missing)age < 68.5 to the right, improve=1.2020, (0 missing)gleason < 5.5 to the left, improve=0.6474, (3 missing)eet < 1.5 to the left, improve=0.1910, (1 missing)Surrogate splits:ploidy splits as LR-, agree=0.917, adj=0.815, (0 split)age < 65.5 to the left, agree=0.617, adj=0.148, (1 split)Node number 3: 85 observations, complexity param=0.02335predicted class=1 expected loss=0.3429 P(node) =0.6341class counts: 40 45probabilities: 0.343 0.657left son=6 (40 obs) right son=7 (45 obs)Primary splits:g2 < 13.2 to the left, improve=1.96800, (6 missing)ploidy splits as LRR, improve=1.84100, (0 missing)age < 56.5 to the right, improve=1.32300, (0 missing)gleason < 8.5 to the left, improve=1.26900, (0 missing)eet < 1.5 to the right, improve=0.09632, (1 missing)Surrogate splits:ploidy splits as LRL, agree=0.962, adj=0.914, (6 split)age < 68.5 to the right, agree=0.608, adj=0.114, (0 split)gleason < 6.5 to the left, agree=0.582, adj=0.057, (0 split)Node number 4: 33 observationspredicted class=0 expected loss=0.05055 P(node) =0.1832class counts: 32 1probabilities: 0.949 0.051Node number 5: 28 observations, complexity param=0.0153predicted class=0 expected loss=0.4053 P(node) =0.1828class counts: 20 8probabilities: 0.595 0.405left son=10 (20 obs) right son=11 (8 obs)Primary splits:gleason < 5.5 to the left, improve=1.1580, (1 missing)age < 67.5 to the right, improve=1.1420, (0 missing)g2 < 14.37 to the right, improve=0.8086, (1 missing)Node number 6: 40 observations, complexity param=0.02335predicted class=1 expected loss=0.4426 P(node) =0.2824class counts: 23 17probabilities: 0.443 0.557left son=12 (7 obs) right son=13 (33 obs)Primary splits:g2 < 11.84 to the right, improve=1.42400, (5 missing)ploidy splits as LLR, improve=1.31000, (0 missing)gleason < 7.5 to the left, improve=0.86340, (0 missing)age < 61.5 to the right, improve=0.57940, (0 missing)eet < 1.5 to the left, improve=0.05363, (0 missing)Surrogate splits:ploidy splits as RLR, agree=0.857, adj=0.286, (5 split)Node number 7: 45 observations, complexity param=0.02275predicted class=1 expected loss=0.2627 P(node) =0.3517class counts: 17 28probabilities: 0.263 0.737left son=14 (22 obs) right son=15 (23 obs)Primary splits:g2 < 17.91 to the right, improve=4.07900, (1 missing)age < 62.5 to the right, improve=0.66100, (0 missing)gleason < 7.5 to the left, improve=0.16170, (0 missing)eet < 1.5 to the right, improve=0.09922, (1 missing)Surrogate splits:age < 61.5 to the right, agree=0.614, adj=0.190, (1 split)eet < 1.5 to the right, agree=0.591, adj=0.143, (0 split)grade < 3.5 to the right, agree=0.545, adj=0.048, (0 split)gleason < 6.5 to the right, agree=0.545, adj=0.048, (0 split)Node number 10: 20 observationspredicted class=0 expected loss=0.2987 P(node) =0.124class counts: 16 4probabilities: 0.701 0.299Node number 11: 8 observationspredicted class=1 expected loss=0.3699 P(node) =0.05878class counts: 4 4probabilities: 0.370 0.630Node number 12: 7 observationspredicted class=0 expected loss=0.2212 P(node) =0.04187class counts: 6 1probabilities: 0.779 0.221Node number 13: 33 observations, complexity param=0.01892predicted class=1 expected loss=0.3841 P(node) =0.2405class counts: 17 16probabilities: 0.384 0.616left son=26 (25 obs) right son=27 (8 obs)Primary splits:g2 < 11 to the left, improve=1.06300, (5 missing)ploidy splits as L-R, improve=0.73950, (0 missing)gleason < 6.5 to the left, improve=0.57220, (0 missing)age < 62.5 to the right, improve=0.13320, (0 missing)eet < 1.5 to the left, improve=0.05492, (0 missing)Surrogate splits:age < 70.5 to the left, agree=0.821, adj=0.286, (5 split)Node number 14: 22 observations, complexity param=0.02275predicted class=0 expected loss=0.4933 P(node) =0.1502class counts: 14 8probabilities: 0.507 0.493left son=28 (15 obs) right son=29 (7 obs)Primary splits:age < 62.5 to the right, improve=0.976400, (0 missing)g2 < 23.48 to the right, improve=0.284300, (1 missing)gleason < 7.5 to the left, improve=0.003765, (0 missing)Surrogate splits:gleason < 6.5 to the right, agree=0.773, adj=0.286, (0 split)Node number 15: 23 observationspredicted class=1 expected loss=0.08092 P(node) =0.2015class counts: 3 20probabilities: 0.081 0.919Node number 26: 25 observations, complexity param=0.01892predicted class=1 expected loss=0.4276 P(node) =0.1779class counts: 14 11probabilities: 0.428 0.572left son=52 (17 obs) right son=53 (8 obs)Primary splits:ploidy splits as L-R, improve=2.157000, (0 missing)gleason < 6.5 to the left, improve=1.972000, (0 missing)age < 61.5 to the right, improve=0.971800, (0 missing)g2 < 7.65 to the right, improve=0.404500, (4 missing)eet < 1.5 to the left, improve=0.002614, (0 missing)Surrogate splits:eet < 1.5 to the right, agree=0.72, adj=0.125, (0 split)g2 < 4.33 to the right, agree=0.72, adj=0.125, (0 split)Node number 27: 8 observationspredicted class=1 expected loss=0.2605 P(node) =0.0626class counts: 3 5probabilities: 0.260 0.740Node number 28: 15 observationspredicted class=0 expected loss=0.3825 P(node) =0.09682class counts: 11 4probabilities: 0.617 0.383Node number 29: 7 observationspredicted class=1 expected loss=0.3057 P(node) =0.05334class counts: 3 4probabilities: 0.306 0.694Node number 52: 17 observationspredicted class=0 expected loss=0.4152 P(node) =0.1115class counts: 12 5probabilities: 0.585 0.415Node number 53: 8 observationspredicted class=1 expected loss=0.1636 P(node) =0.06643class counts: 2 6probabilities: 0.164 0.836> #> # Fit a classification model to the car data.> # Now, since Reliability is an ordered category, this model doesn't> # make a lot of statistical sense, but it does test out some> # areas of the code that nothing else does> #> xcar <- rep(1:8, length=nrow(cu.summary))> carfit <- rpart(Reliability ~ Price + Country + Mileage + Type,+ method='class', data=cu.summary, xval=xcar)>> summary(carfit)Call:rpart(formula = Reliability ~ Price + Country + Mileage + Type,data = cu.summary, method = "class", xval = xcar)n=85 (32 observations deleted due to missingness)CP nsplit rel error xerror xstd1 0.30508 0 1.0000 1.0000 0.072002 0.08475 1 0.6949 0.6949 0.078083 0.05085 2 0.6102 0.7119 0.078134 0.03390 3 0.5593 0.6780 0.078005 0.01000 4 0.5254 0.6780 0.07800Variable importanceCountry Type Price66 22 12Node number 1: 85 observations, complexity param=0.3051predicted class=average expected loss=0.6941 P(node) =1class counts: 18 12 26 8 21probabilities: 0.212 0.141 0.306 0.094 0.247left son=2 (58 obs) right son=3 (27 obs)Primary splits:Country splits as ---LRRLLLL, improve=15.220, (0 missing)Type splits as RLLRLL, improve= 4.288, (0 missing)Price < 11970 to the right, improve= 3.200, (0 missing)Mileage < 24.5 to the left, improve= 2.476, (36 missing)Node number 2: 58 observations, complexity param=0.08475predicted class=average expected loss=0.6034 P(node) =0.6824class counts: 18 12 23 5 0probabilities: 0.310 0.207 0.397 0.086 0.000left son=4 (9 obs) right son=5 (49 obs)Primary splits:Type splits as RRRRLR, improve=3.187, (0 missing)Price < 11230 to the left, improve=2.564, (0 missing)Mileage < 24.5 to the left, improve=1.802, (30 missing)Country splits as ---L--RLRL, improve=1.329, (0 missing)Node number 3: 27 observationspredicted class=Much better expected loss=0.2222 P(node) =0.3176class counts: 0 0 3 3 21probabilities: 0.000 0.000 0.111 0.111 0.778Node number 4: 9 observationspredicted class=Much worse expected loss=0.2222 P(node) =0.1059class counts: 7 0 2 0 0probabilities: 0.778 0.000 0.222 0.000 0.000Node number 5: 49 observations, complexity param=0.05085predicted class=average expected loss=0.5714 P(node) =0.5765class counts: 11 12 21 5 0probabilities: 0.224 0.245 0.429 0.102 0.000left son=10 (27 obs) right son=11 (22 obs)Primary splits:Type splits as RLLR-L, improve=2.880, (0 missing)Mileage < 24.5 to the left, improve=2.500, (25 missing)Price < 11470 to the right, improve=2.424, (0 missing)Country splits as ---R--LRLR, improve=1.027, (0 missing)Surrogate splits:Price < 11470 to the right, agree=0.898, adj=0.773, (0 split)Country splits as ---R--RRRL, agree=0.755, adj=0.455, (0 split)Node number 10: 27 observationspredicted class=average expected loss=0.4074 P(node) =0.3176class counts: 7 4 16 0 0probabilities: 0.259 0.148 0.593 0.000 0.000Node number 11: 22 observations, complexity param=0.0339predicted class=worse expected loss=0.6364 P(node) =0.2588class counts: 4 8 5 5 0probabilities: 0.182 0.364 0.227 0.227 0.000left son=22 (14 obs) right son=23 (8 obs)Primary splits:Country splits as ---R--LRRL, improve=1.5190, (0 missing)Price < 8646 to the left, improve=1.2720, (0 missing)Type splits as L--R--, improve=0.1909, (0 missing)Surrogate splits:Price < 13970 to the left, agree=0.864, adj=0.625, (0 split)Node number 22: 14 observationspredicted class=worse expected loss=0.5714 P(node) =0.1647class counts: 4 6 1 3 0probabilities: 0.286 0.429 0.071 0.214 0.000Node number 23: 8 observationspredicted class=average expected loss=0.5 P(node) =0.09412class counts: 0 2 4 2 0probabilities: 0.000 0.250 0.500 0.250 0.000>>> proc.time()user system elapsed0.480 0.039 0.504