Thread Subject: Give me a Regression Problem

On Jul 19, 4:22=A0pm, Greg Heath <he...@alumni.brown.edu> wrote:
> On Jul 19, 9:22 am, baldrick <philbrier...@hotmail.com> wrote:
-----SNIP
> > Just been doing a bit more messing with the concrete data and it
> > appears that drying concrete has a bit of a half life. If you take the
> > natural logarithm of the age and use this then you should be able to
> > get an r^2 of 0.825 using linear regression. Should save you a few
> > neurons.
>
> Nice !

-----SNIP

> Hmm ... how many neurons can be saved using log10(age)?

Sorry I can't post plots.
However, if you compare the "age" summary stats
( H =3D 0 =3D=3D> no hidden layer =3D=3D> iterative linear regression )

 H min median mean stdv max
 0 0.572 0.658 0.655 0.046 0.726
 1 0.603 0.679 0.683 0.042 0.746
 2 0.764 0.820 0.816 0.030 0.850
 3 0.831 0.865 0.864 0.019 0.891
 4 0.834 0.873 0.870 0.021 0.898
 5 0.843 0.878 0.877 0.018 0.902
 6 0.850 0.890 0.886 0.018 0.913
 7 0.863 0.887 0.889 0.020 0.921
 8 0.880 0.900 0.896 0.013 0.915
 9 0.864 0.892 0.898 0.021 0.936
10 0.790 0.900 0.894 0.041 0.930
20 0.827 0.924 0.909 0.039 0.945
30 0.860 0.933 0.925 0.029 0.962
40 0.900 0.921 0.924 0.019 0.950
50 0.852 0.930 0.924 0.029 0.952
60 0.831 0.933 0.920 0.036 0.953
70 0.855 0.934 0.928 0.029 0.957

with the "log10(age)" summary stats,

 H min median mean stdv max
 0 0.698 0.746 0.743 0.039 0.819
 1 0.767 0.823 0.822 0.034 0.869
 2 0.786 0.848 0.849 0.030 0.893
 3 0.800 0.858 0.857 0.035 0.905
 4 0.855 0.888 0.880 0.019 0.909
 5 0.831 0.893 0.887 0.028 0.926
 6 0.852 0.892 0.892 0.023 0.935
 7 0.852 0.894 0.896 0.022 0.930
 8 0.884 0.902 0.908 0.021 0.941
 9 0.861 0.909 0.903 0.021 0.933
10 0.866 0.911 0.908 0.025 0.941
20 0.906 0.928 0.929 0.015 0.954
30 0.831 0.931 0.921 0.038 0.955
40 0.891 0.943 0.934 0.023 0.954
50 0.894 0.931 0.932 0.017 0.961
60 0.915 0.941 0.939 0.018 0.965
70 0.833 0.942 0.930 0.036 0.958

you will conclude that the significant advantage dissapears by
the time H =3D 3.

Nevertheless, using log10(age) with quadratic regression
should reveal all of the important effects.

Hope this helps.

Greg