I have read that it is good to normalize data before training a Neural Nework and i have a perfect result with normalizing database as a normal data. I normalize inputs/targets to fall in range [-1 1] using premnmx / postmnmx.
All work is done on the script and I want to turn it into a simulink with normal data.
Hello,
This is a good question. Please consider that the initial normalization process depends on the transfer function you have implemented in the NN. If the operating range of the transfer function is from 1 to 0 or -1 to 1 or any other, you should normalize your database accordingly. If you don't do that, you are reducing the operating range of the NN designed and increasing the estimating error. This is the most important point, in my opinion the tool used to develop the neural network is not so important.
Please take a look to the chapter attached. I hope you will find it useful.
Cheers,
José.
Dear Khadija
You can keep the Normalization Script as is (a MatLab function) and then use the "Matlab-Function" block in Simulink to call the script at every time step. It is very easy to use.
avoid simulink, it will curb your understanding. Best is to keep working in .m file
For neural network, all the time you should scale the data (at least input data) in .m file of Simulink. And as Ahmed says, call the Script file using Matlab-Function Block in Simulink.
Best wishes
Hello,
This is a good question. Please consider that the initial normalization process depends on the transfer function you have implemented in the NN. If the operating range of the transfer function is from 1 to 0 or -1 to 1 or any other, you should normalize your database accordingly. If you don't do that, you are reducing the operating range of the NN designed and increasing the estimating error. This is the most important point, in my opinion the tool used to develop the neural network is not so important.
Please take a look to the chapter attached. I hope you will find it useful.
Cheers,
José.
Hello,
This is a good question. Please consider that the initial normalization process depends on the transfer function you have implemented in the NN. If the operating range of the transfer function is from 1 to 0 or -1 to 1 or any other, you should normalize your database accordingly. If you don't do that, you are reducing the operating range of the NN designed and increasing the estimating error. This is the most important point, in my opinion the tool used to develop the neural network is not so important.
Please take a look to the chapter attached. I hope you will find it useful.
Cheers,
José.
Hello,
This is a good question. Please consider that the initial normalization process depends on the transfer function you have implemented in the NN. If the operating range of the transfer function is from 1 to 0 or -1 to 1 or any other, you should normalize your database accordingly. If you don't do that, you are reducing the operating range of the NN designed and increasing the estimating error. This is the most important point, in my opinion the tool used to develop the neural network is not so important.
Please take a look to the chapter attached. I hope you will find it useful.
Cheers,
José.
Hi. It easy to normalized your data and the base of NN is normality. you can use of programs such az (spss, minitab and matlab). you do normality test. if your data was normal that is good , use in NN. but are not normal you can normal them . I can help you if need.
thanks and good luck
Save the trained network with the normalized data
save net
or
save('NetName', 'net', '-mat');
and then in Simulink ,you can load the saved net (Use function block to call the load net)
load net
or
load NetName
http://www.mathworks.com/help/imaq/acquiring-image-and-skeletal-data-using-the-kinect.html
The noise in ANN with normal data is considerably high, unless your data range is limited. I believe Matlab normalized input before running the function.
I would suggest you to normalize your data before feeding it to Matlab. This is important as it eliminates noise. I did mine in excel using z-score normalization, which is quite simple to implement. However, your normalization technique will depend on your dataset and your approach.
Are you asking for advice on data-normalization or about how to convert a Matlab script into a Simulink simulation?
Advice on normalization:
I would not recommend basing normalization on max/min in training data, like premnmx does, because max and min are dominated by a few outliers who might not give a good representation of actual or typical limits. This is the noise-issue mentioned above.
The rescaling algorithm is simple: pn = (p-minp) * 2 / (maxp -minp) - 1;
Or equivalently: pn = (p - middle) / range;
(Here p is a raw data-point, and pn is a normalized one. The aim is that [-1
- Badges
- Science topic