Not expected weights size

I'm training a neural network with 25344 neurons in input layer, 64 neurons in hidden layer and 4 neurons in output layer. So i expect to get something like 25345x64 Theta1 and 65x4 Theta2, but getwb() returns me a 1622018x1 sized vector. What should i do to get Theta1 and Theta2 files of right sizes?

[Greg Heath]

The nodes in the input layer are not neurons; they are fan-in units.

I have never heard of the notation Theta1 and Theta2. The standard MATLAB notations are

 IW = net.IW;
  b = net.b;
 LW = net.LW;

Read the results of both the help and doc documentation commands on each of the functions

 configure, getwb, setwb, formwb and separatewb

e.g.

 help configure 
 doc formwb

Hope this helps.

Greg

Answer 1

[Greg Heath]

1. I have found the "I"nput and "O"utput target vector dimensions for the MATLAB NN EXAMPLE DATA SETS using the help and doc commands, i.e.,

 help nndatasets
 doc nndatasets

I then posted them in the NEWSREADER threads

 www.mathworks.com/matlabcentral/newsreader/...
.../view_thread/337914  % Regression
.../view_thread/338208  % Timeseries
 .../view_thread/339984  % Classification

2. Read both the help and doc documentations for functions

 configure, getwb, setwb, formwb and separatewb

3. For a feedforward net with an I-H-O "I"nput-"H"idden-"O"utput node topology, the TOTAL number of weights is

 Nw = (I+1)*H+(H+1)*O

where the INDIVIDUAL numbers are

 IW:  input-to-hidden    I*H
 B1:  bias-to-hidden     1*H
 LW: hidden-to-output    H*O
 B2:  bias-to-output     1*O

4. For regression examples 4 and/or 6 and/or 7, i.e.,

 [ x4, t4 ] = building_dataset % Building Energy dataset
 [ x6, t6 ] = cho_dataset;     % Cholesterol dataset
 [ x7, t7 ] = engine_dataset;  % Engine behavior dataset

use the following command to create the default configuration for the regression net

 net = fitnet;

5. Apply the functions in part 2 to as many of the datasets in part 4 as you need in order to understand how they work with respect to one another

6. Post if you have any more questions

 Hope this helps.

 *Thank you for formally accepting my answer*

 Greg