Process Help |
Process Name |
Menu Path |
Link to Command Table |
RNLM |
Sample Analysis ribbon | Geochemical Processes | R Mode Non Linear Mapping |
Introduction
R - mode non linear mapping groups together elements using the angular distances calculated from the Pearsson correlation coefficient matrix.
Note the difference with Q mode non linear mapping which clusters samples on the basis of composition.
How to use
A two dimensional view of the element clusters, or the scores NLM-X versus NLM-Y, is calculated to represent them as they would appear in multi-dimensional space on the basis of element similarity between pairs of samples. All input data is standardized prior to the calculation of the correlation matrix and the linear mapping output scores are normalized prior to plotting.
R - mode non linear mapping, when compared with R - mode factor analysis, is more sophisticated and does not distort or sub-divide the element or parameter clusters. Non linear mapping will give more separable clusters than other hierarchical techniques such as R - mode cluster analysis.
File Handling
The input file &(IN) must have a separate identifier field (*SAMPID). The output file &(SCORES) contains three parameters, FIELD the field identifier, NLM-X and NLM-Y the output scores for plotting the multi-dimensional inter-element distances as a two dimensional flat plane. Results can be displayed using QUIG, PLOTDA or PLOTAN.
Iteration Procedure
In order to present a two dimensional view of multi-dimensional space with minimum distortion of the sample clusters, the calculated mapping error has to be minimized by an iterative method (steepest descent). This is controlled by the @CONVLIM parameter, that is the minimum difference allowed in the mapping error between iterations, @MAXIT, the maximum number of iterations permitted and the @MAGIC parameter which specifies the stepping function used for each iteration. If the stepping function @MAGIC is decreased, the number of iterations is increased with an obvious time penalty on the length of the calculation. The value used must be taken into account when there are a large number of samples. However the results can be more stable.
Files, Fields and Parameters
Input Files
Name |
Description |
I/O Status |
Required |
Type |
IN |
Input file. |
Input |
Yes |
Undefined |
Output Files
Name |
I/O Status |
Required |
Type |
Description |
SCORES |
Output |
No |
Undefined |
Optional output file for principal component scores. |
Fields
Name |
Description |
Source |
Required |
Type |
Default |
SAMPID |
Field containing sample identification |
IN |
Yes |
Any |
Undefined |
F1 |
First field to be used. No fields specified means all. |
IN |
No |
Numeric |
Undefined |
F2 |
Second field to be used. |
IN |
No |
Numeric |
Undefined |
F3 |
Third field to be used. |
IN |
No |
Numeric |
Undefined |
F4 |
Fourth field to be used. |
IN |
No |
Numeric |
Undefined |
F5 |
Fifth field to be used. |
IN |
No |
Numeric |
Undefined |
F6 |
Sixth field to be used. |
IN |
No |
Numeric |
Undefined |
F7 |
Seventh field to be used. |
IN |
No |
Numeric |
Undefined |
F8 |
Eighth field to be used. |
IN |
No |
Numeric |
Undefined |
F9 |
Ninth field to be used. |
IN |
No |
Numeric |
Undefined |
F10 |
Tenth field to be used. |
IN |
No |
Numeric |
Undefined |
Parameters
Name |
Description |
Required |
Default |
Range |
Values |
||||
CONVLIM |
Convergence limit (0.0001) |
No |
0.0001 |
Undefined |
Undefined |
||||
MAGIC |
Convergence [magic] factor (0.35) |
No |
0.35 |
Undefined |
Undefined |
||||
MAXIT |
Maximum number of iterations (100) |
No |
100 |
Undefined |
Undefined |
||||
|
|
No |
0 |
0,1 |
0,1 |
Notes
There is a restriction of 2000 samples. Note, as the process is iterative, it may take some time to calculate the results (30 minutes).
Example
!RNLM
|
&IN(AEG2),@SCORES(RSCORES),@SAMPID='ID', |
Error and Warning Messages
Message |
Description |
Solution |
*** Error *** No numeric fields on file. >>> ERROR 122 <<< ( fileno.) IN RNLM |
|
|
|
|
|
*** Error *** Compulsory file, field or parameter missing. >>> ERROR 120 <<< ( fileno.) IN RNLM |
|
|
|
|
|