Weight spreadsheet

Have specific questions about how to work with certain MemBrain features? Not sure about which checkbox in MemBrain has which effects? Want to know if a certain functionality is available in MemBrain or not?

Your questions will be appreciated here!
Post Reply
alexeymosc
Posts: 5
Joined: Tue 30. Nov 2010, 16:34

Weight spreadsheet

Post by alexeymosc » Tue 30. Nov 2010, 16:39

Hello!

I use MemBrain. Thank you! I like this program. Could you kindly advise if there is a way to get something like a spreadsheet with all the synaptic weights for further analysis? As I can tell there is only an option of displaying weights on the synapses. Is it possible to export a document with all weights?

Thank you,
Alexey

User avatar
Admin
Site Admin
Posts: 438
Joined: Sun 16. Nov 2008, 18:21

Re: Weight spreadsheet

Post by Admin » Tue 30. Nov 2010, 19:08

alexeymosc wrote:Is it possible to export a document with all weights?
Yes, there is:
You can use <File><Export...> to export the whole net into a sectioned CSV file. The feature also lets you select the items you want to export.
Please take into considerations that not only the links define your net but also the activation thresholds of the neurons. You can look at these like links to so called 'ON neurons'.

Regards,
Thomas Jetter

alexeymosc
Posts: 5
Joined: Tue 30. Nov 2010, 16:34

Re: Weight spreadsheet

Post by alexeymosc » Wed 1. Dec 2010, 11:35

Dear Thomas,

Thank you for this information. Anyway the program is great and much fun (as compared to some other software available). The help files are good also, and very clear.

As an idea how to improve the program I would suggest the option to display both training and testing (and possibly more) net errors during net training.

Alexey

alexeymosc
Posts: 5
Joined: Tue 30. Nov 2010, 16:34

Re: Weight spreadsheet

Post by alexeymosc » Wed 1. Dec 2010, 21:22

Please take into considerations that not only the links define your net but also the activation thresholds of the neurons.
Dear Thomas, do you mean by activation thresholds the 'biases' which are often called so in neural network literature? I mean a linear component to the sum of weighted outputs.

User avatar
Admin
Site Admin
Posts: 438
Joined: Sun 16. Nov 2008, 18:21

Re: Weight spreadsheet

Post by Admin » Wed 1. Dec 2010, 22:27

alexeymosc wrote:do you mean by activation thresholds the 'biases' which are often called so in neural network literature? I mean a linear component to the sum of weighted outputs.
Yes and no:
I mean the 'biases' but they are a component to the INPUT sum of each neuron. To put it more precisely: The threshold is a negative summand to the input sum, e.g. an activation threshold of 0.4 means that the sum of inputs is reduced by the value 0.4 before subjected to the activation function.

An alternative interpretation of the threshold is a negative link to the output of a so-called 'ON-neuron', i.e. to a neuron whose output is permanently 1. Some neural network simulators thus don't have 'Activation Thresholds' but links to explicit ON-neurons instead. Mathematically, they are of identical effect.


Regards,
Thomas
Thomas Jetter

alexeymosc
Posts: 5
Joined: Tue 30. Nov 2010, 16:34

Re: Weight spreadsheet

Post by alexeymosc » Thu 2. Dec 2010, 15:01

Dear Thomas,

Thanks for explanation. It is clear.

Alexey

Post Reply