What kind or Result analysis can be done to analyse the characteristics of Restricted boltzmann machine? I have used Mean square error value is one of the parameter to analyse.I also noticed that every time i run a design i get different Mean square error value.Not able to identify the reason .
What would be other parameters used to analyse?
 A: There is a very nice paper by Geoffry Hinton in which he gives practical advise on how to train RBMs: http://www.cs.toronto.edu/~hinton/absps/guideTR.pdf
If you are training your RBM on images you can plot the weights from the hidden to the visible units. To give an example: As an input you have a 16 x 16 pixel image (in total 256 pixels) thus you got 256 visible units. Suppose you have 50 hidden units than from each hidden unit, there are 256 weights, one to each visible unit. You plot each weight from a given hidden unit to a visible unit according to the position the visible unit has in your image (you will end up with 50 figures, one for each hidden unit). If we are looking at hidden unit 1, weight 1,1 (connecting hidden unit 1 with visible unit 1), that weight has to fill the pixel at position 1 x 1 in image 1). If your training works, you will see patterns in your images. 
What I forgot to mention is, that these images are usually greyscaled. 
Furthermore you can also approximate the likelihood value and/or the free energy. (More stuff can be found in the paper)
And of course you will most unlikely end up with the same value (think of the sampling steps you are taking during each weight update).
