R : Copyright 2004, The R Foundation for Statistical Computing
Version 2.0.0  (2004-10-04), ISBN 3-900051-07-0

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 a HTML browser interface to help.
Type 'q()' to quit R.

> invisible(options(echo = TRUE))
> source("mlp.r")
> 
> trnx = as.matrix(read.table("a2trnx",head=F))
> tstx = as.matrix(read.table("a2tstx",head=F))
> 
> trny = scan("a2trny")
Read 1708 items
> tsty = scan("a2tsty")
Read 2562 items
> 
> # Try logistic regression.
> 
> m = glm (trny ~ trnx, family = "binomial")
> 
> print(summary(m))

Call:
glm(formula = trny ~ trnx, family = "binomial")

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-2.9128  -0.4880   0.2613   0.5866   2.9459  

Coefficients:
            Estimate Std. Error z value Pr(>|z|)    
(Intercept) -3.05071    1.26836  -2.405 0.016161 *  
trnxV1      -0.97554    0.19350  -5.042 4.62e-07 ***
trnxV2      -1.28711    0.23445  -5.490 4.02e-08 ***
trnxV3      -0.01075    0.09407  -0.114 0.909025    
trnxV4       0.42009    0.23313   1.802 0.071549 .  
trnxV5       0.26227    0.16644   1.576 0.115090    
trnxV6      -0.04156    0.02763  -1.504 0.132585    
trnxV7       0.57519    0.18703   3.075 0.002103 ** 
trnxV8      -3.54666    0.27940 -12.694  < 2e-16 ***
trnxV9       0.06427    0.18545   0.347 0.728912    
trnxV10     -0.19697    0.11305  -1.742 0.081440 .  
trnxV11      0.10247    0.09171   1.117 0.263825    
trnxV12      0.07966    0.15494   0.514 0.607143    
trnxV13     -0.20463    0.13087  -1.564 0.117907    
trnxV14     -1.19928    0.27087  -4.427 9.53e-06 ***
trnxV15      0.74670    0.20644   3.617 0.000298 ***
trnxV16      5.60564    0.46877  11.958  < 2e-16 ***
trnxV17     -0.31754    0.24613  -1.290 0.197010    
trnxV18     -0.07661    0.28577  -0.268 0.788643    
trnxV19      2.47395    0.29163   8.483  < 2e-16 ***
trnxV20     -3.22532    0.30526 -10.566  < 2e-16 ***
trnxV21      0.96260    0.36075   2.668 0.007623 ** 
trnxV22      0.86491    0.33945   2.548 0.010835 *  
trnxV23      1.75445    0.34105   5.144 2.69e-07 ***
---
Signif. codes:  0 `***' 0.001 `**' 0.01 `*' 0.05 `.' 0.1 ` ' 1 

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 2285.2  on 1707  degrees of freedom
Residual deviance: 1320.6  on 1684  degrees of freedom
AIC: 1368.6

Number of Fisher Scoring iterations: 5

> 
> p1.logistic = 1 / (1 + exp(-(coef(m)[1] + tstx %*% coef(m)[2:(ncol(tstx)+1)])))
> 
> cat("Error rate with logistic regression:\n");
Error rate with logistic regression:
> print(mean(tsty != (p1.logistic>0.5)))
[1] 0.1791569
> 
> val.set = 1:427
> est.set = 428:1708
> 
> postscript("ass2-plts.ps",horiz=T)
> par(mfrow=c(2,4))
> 
> # Try MLP with varous learning rates.
> 
> iters = 
+    c (   60000,    30000,   20000,  15000,  10000,    60000,    60000,    60000)
> eta1  = 
+    c (0.000012, 0.000025, 0.00005, 0.0001, 0.0002, 0.000012, 0.000012, 0.000020)
> eta2  = 
+    c (0.000012, 0.000025, 0.00005, 0.0001, 0.0002, 0.000020, 0.000030, 0.000012)
> 
> for (i in 1:length(iters))
+ {
+   set.seed(123)
+ 
+   params = mlp.cv (trny, trnx, eta1[i], eta2[i], val.set, iters=iters[i], q=15,
+                    cv.plot=T)
+   title(paste("eta1 =",eta1[i],"   eta2 =",eta2[i]))
+   p1.mlp = mlp.predict(tstx,params)
+   cat("Error rate with mlp using eta1 =",eta1[i],"and eta2 =",eta2[i],":\n")
+   print(mean(tsty != (p1.mlp>0.5)))
+ }
Error rate with mlp using eta1 = 1.2e-05 and eta2 = 1.2e-05 :
[1] 0.1436378
Error rate with mlp using eta1 = 2.5e-05 and eta2 = 2.5e-05 :
[1] 0.1436378
Error rate with mlp using eta1 = 5e-05 and eta2 = 5e-05 :
[1] 0.1444184
Error rate with mlp using eta1 = 1e-04 and eta2 = 1e-04 :
[1] 0.1506635
Error rate with mlp using eta1 = 2e-04 and eta2 = 2e-04 :
[1] 0.1807182
Error rate with mlp using eta1 = 1.2e-05 and eta2 = 2e-05 :
[1] 0.1405152
Error rate with mlp using eta1 = 1.2e-05 and eta2 = 3e-05 :
[1] 0.1412959
Error rate with mlp using eta1 = 2e-05 and eta2 = 1.2e-05 :
[1] 0.1451991
> proc.time()
[1] 8309.61   29.31 9546.92    0.01    0.00
>