Next Previous Table of Contents
The general procedure for analyzing account data is as follows:
These steps and other optional procedures are explaned below.
Account data is loaded from menu File/Open.
The data file contains a number of monthly account balances from a continuous time period. The last 12 months of the data are considered as test data, and all the data before that is used as training data.
The training data is used to predict what the account balaces in the last year should be. The accountant compares that prediction to the actual balances, the test data in the file. If there is a great difference in a certain balance, the accountant may choose to pay more attention to it.
The file format must be exactly correct, as explained here. The immediately first row is a header that contains the names of the accounts, separated by tab characters. This first column of the first row is empty, followed by a single tab character.
The rest of the rows are data rows. The columns are separated by a single tab character. First column of each data row is time label. Two first numbers are the year and two latter numbers are the month. The year is always given with two digits (Y2k problems won't occur before 2030).
Below is an example from the beginning of an account data file.
Net sales Materials Person costs Gross margin Administration Total indirect Operating profit Receivables Trade debts
9001 4769 -974 -1041 2754 -297 -1713 1041 8184 -6891
9002 3794 -874 -717 2203 -112 -1074 1129 8612 -8334
9003 4846 -1315 -938 2593 -387 -2191 402 8613 -8153
9004 4452 -1564 -922 1966 -136 -1582 384 9204 -8071
9005 4665 -1283 -702 2680 -424 -1995 685 9492 -8704
9006 4420 -1301 -738 2381 -316 -2041 340 9818 -8541
9007 871 -242 -193 436 -92 -549 -113 9513 -8110
9008 5561 -1451 -1046 3064 -355 -2274 790 9558 -8023
9009 3666 -1145 -799 1722 -268 -1731 -9 9624 -7983
9010 4179 -1043 -777 2359 -216 -1832 527 9615 -8091
9011 3924 -1013 -1077 1834 -295 -1305 529 9615 -8161
9012 2572 -792 -814 966 -290 -2166 -1200 8609 -7794
9101 3584 -723 -840 2021 -343 -1824 197 8866 -7693
9102 3151 -707 -568 1876 -436 -1517 359 7737 -7700
9103 3703 -825 -652 2226 -285 -1947 279 8820 -7299
...
Note that the first row may seem to continue on the second row, but that because of word wrap in this document. The actual data file must not be word-wrapped.
The final row containing numbers must have a trailing newline, but there must be nothing after that.
Most spreadsheet softwares support saving tables in tab spaced format.
Gnumeric supports tab spaced values. Use menu File/Save As and save the file as "Simple Text". The problem is that GNumeric puts the name of the sheet as the first row. You must remove it with a text editor.
KSpread supports only comma spaced values (CSV). Choose
menu item File/Save As, select the CSV format, and save the
file with name datafile.csv
. Then open a shell window, change
to appropriate directory, and issue following
$ perl -pe 's/,/\t/g' datafile.csv >datafile-tab.csv
This changes the comma separators to tab characters. Check that the file format is correct.
Budget data files follow exactly the same format as the actual data file. The column names in the first row are ignored. The number and order of the data columns must be exactly same as in training data file.
Changing the neural network training options shouldn't be done by the accountant in routine operation. The options are mostly for research purposes.
Next Previous Table of Contents