austenite data

original data: database3.

Data as used by CALJ with Tpros: in directories 12d and 22d. Please see directory GPmanual.

Tue Aug 25 13:43:29 1998
From: calj@mpia-hd.mpg.de (Coryn Bailer-Jones)
I have slightly re-written the gpaus manual (on the application of Tpros to the Austenite data) which I wrote before I left. I have also collected together all of the files required for this problem into a single place, without any redundant files, which is considerably neater than before. The whole lot I have put into /home/calj/data/gpaus.tar

This tar file can be obtained from here (gz).

Old info about austenite
I edited a few lines where the last two entries appeared to have run together, by making 923.xy.0000 into 923.x y.0000

But that was wrong. Harry said 882 923.50.0000 is actually

882 923 50.0000
Ac1 Ac3 Heating rate
OK. I patched these oddities up.

789 lines, 25 entries.

I deleted the initial entry (line number)

The remaining quantities are: 1 C 2 Si 3 Mn 4 S 5 P 6 Cu 7 Ni 8 Cr 9 Mo 10 Nb 11 V 12 Ti 13 Al 14 B 15 W 16 As 17 Sn 18 Zr 19 Co 20 N 21 O TARGET 1 Ac1 TARGET 2 Ac3 22 Heating Rate

I used my rescaling program

~/bin/maxmin.p topskip=1 < tmp > rescaler.p

to write a rescaler.p to rescale these. The program rescaler.p contains a record of the min and max values encountered. These agree with the paper. tmp0 is the raw file with header removed. To make a fully rescaled data file:

 rescaler.p tmp0 > rescaled
This is 789 lines, 24 entries. I have pulled the two target columns out into a file called targets. And split them into target files called ac1.ta and ac3.ta. NB, none of these files have been reordered in any way

Training/test division

The training set should be a random selection of 394 examples. The test set is the other 395 (or 394) ---

For no particular reason, the very last line is to be omitted if the experiments are to match Harry's exactly. But this line was only omitted to make the total number of examples even, so it might as well be left in.


David MacKay <mackay@mrao.cam.ac.uk>
Last modified: Fri Aug 28 10:44:15 1998