Self Organization Map (SOM)

HSP Application note #45

2010.7.26

HSPiP Team Senior Developer, Dr. Hiroshi Yamamoto

SOM: Self Organization Map Neural Network

The 2D Map of "Smilar vector map to similar 2D position".

If we split dH to dHdo, dHac then HSP vector become 4 dimensions.
That is the very bad news for user. Because the Sphere and GUI can not expand to 4 dimensions.
So we start to develop SOM program to check vector similarity.

Pirika JAVA Demo Applet calculate SOM. SOMDemo is available here.

Hansen Solubility Parameters (HSP)

Hansen Solubility Parameters(HSP) were developed by Charles M. Hansen as a way of predicting if one material will dissolve in another and form a solution. They are based on the idea that "like dissolves like" where one molecule is defined as being 'like' another if it bonds to itself in a similar way.
Specifically, each molecule is given three Hansen parameters, each generally measured in MPa0.5:
dD:The energy from dispersion bonds between molecules
dP:The energy from dipolar intermolecular force between molecules
dH:The energy from hydrogen bonds between molecules.
These three parameters can be treated as Vector for a point in three dimensions also known as the Hansen space. The nearer two molecules HSP Vector are in this three dimensional space, the more likely they are to dissolve into each other.

What can perhaps be surprising is that one can assign HSP to so many different things. Gases like carbon dioxide, solids like carbon-60, sugar, and biological materials like human skin, depot fat, DNA, and even some proteins all have HSP. The list can be continued with drugs, polymers, plasticizers, and in fact any organic material and even many inorganic materials like salts. The only requirement for an experimental confirmation is that the material must behave differently in a sufficient number of test solvents upon contact.

Pirika JAVA Demo Applet calculate HSP. HSPLight is available here.
Please refer to e-Book of HSPiP if you want know more about HSP.
About the Power Tools that handle HSP more effectively.

You just input HSP and push start button. If the movement is converged, push Stop button.
If you click colored area, you will get HSP value at that point.
Red numbers are HCode for typical solvents. You can get solvent information from HSPiP.

This demo version use only 3D HSP. Adding Volume, Donor/Acceptor information, SOM become very powerful tool.

SOM: Self Organization Map Neural Network
(Map N Dimensional vectors to 2D)

What we want to do is Map 1 Vector to 2D plane so as to "Smilar vector map to similar 2D position".

We have many vectors like,

Vec1 [dD1,dP1,dH1,Vol1,.....]
Vec2 [dD2,dP2,dH2,Vol2,.....]
Vec3 [dD3,dP3,dH3,Vol3,.....]

Vecm [dDm,dPm,dHm,Volm,.....]

At first, put random vectors on the 2D plane.

VecX

Then search most similar vector (Winner) in 2D plane with first vector.

For Winner and around winner, move a little toward input vector.

Repeat 1-m vectors.

And repeat again and agian.

Finally, you will get the 2D Map of "Smilar vector map to similar 2D position".

We always use Euclid Distance for SOM, but for this purpose, I add 4.0 before dD term.
And I want to suggest to use square root Volume for SOM.

HSP Distance

To calculate the distance (Ra) between Hansen parameters in Hansen space the following formula is used:

HSP distance(Ra)={4*(dD1-dD2)2 + (dP1-dP2)2 +(dH1-dH2)2 }0.5

You can copy the result and paste it spreadsheet.

You can easily find out similarity of solvents with SOM.

Adding SOM into HSPiP is future plan.
I do not know when.
If I got a lot of feed back, I will set much higher priority.

Hansen Solubility Parameters(HSP) Application Notes

HSP Application note #45

SOM: Self Organization Map Neural Network

Hansen Solubility Parameters (HSP)

SOM: Self Organization Map Neural Network (Map N Dimensional vectors to 2D)

HSP Distance

SOM: Self Organization Map Neural Network
(Map N Dimensional vectors to 2D)