I am trying to compare response surface models with those of artificial networks as current researches in this area tend to favour neural network modeling.
Hi Maurice, I use QNet 2000 for windows (Vesta software) that I bought some years ago. It was quite cheap ($50 USD, i guess) it's fast, easy to use for feedforward backpropagation ANN's. It presents useful features as early-stopping, lets you choose the networks topology, the connections and transger functions, etc. You may integrate the libraries with other software. You may try to find the company website and check the current price.
Hope it helped. Regards Luis
Dear Maurice,
MATLAB software can be a good choice. Neural Network Toolbox in MATLAB provides algorithms, functions, and apps to create, train, visualize, and simulate neural networks. You can perform classification, regression, clustering, dimensionality reduction, time-series forecasting, and dynamic system modeling and control.
Good Luck.
http://www.mathworks.com/products/neural-network/
Thank you Luis and Soheila, I used SPSS neural networks but i don't feel good about the results I got. Could you please comment on this. Thanks again
Dear Maurice Ekpenyong,
Would you explain more about your question?! The attached file maybe helpful.
Regards.
@Maurice Ekpenyong: The SPSS not providing you with the desired results could be attributed to the kind of training algorithm you are using. For example, it is not always the case that Levenberg Marquardt's algorithm, the most widely used could be better for the learning process. Sometimes you have to change and apply other learning algorithms and you will see improvement. You also have to check whether you are able to normalize your data within the SPSS environment.
I have not used the SPSS but I will suggest you use MATLAB because I think it will be more flexible than the SPSS. For the SPSS you are using only what has been provided but the MATLAB you can try other algorithms.
MATLAB with neural networks tool box is important for your research. Qualnet is another software and it is user friendly (because it is ICON based). No need to write the program.
by
R.Suresh Babu
If you familiar with Matlab, then Matlab Neural Network Toolbox is better choice for you. It is easy to use and you can customize you network easily.
Dear all, Matlab is clearly more powerful than the QNet software I suggested in my previous answer. However if you take into account both the cost of buying matlab with NN toolbox and its learning curve, there is no doubt that QNet is by far "the best option" (please don´t take this sentence out of the context). SPSS is ok and there are lots of Universities that own a site license (as mine does).
The problem with not getting the same results (whatever that means - are you referring to the predicted values) may be due to so many reasons that it is quite difficult to state which one(s) is the "culprit".It may depends on the NN topology/structure, on the training, test and validation data sets, on the learning parameters (the algorithm as Yao pointed out but more important are the algorithm parameters like learning rates etc, early stopping...,you name it).
Like in many NN's you need to train them starting with different initial estimates (usually random), using different subsets of your full data set to train/validate, etc, and check if it converges to the same solution.
As Soheila pointed out you should clarify your question (preferably with example(s)).
Hope it helps.
Best regards
Luis
PS: It's common practice to upvote or downvote answers depending if you find them useful or if they are wrong or are misleading. Please kindly adopt this practice.
Dear all,
I completely agree that for really useful assistance, we need to know more about your problem; it may even be so that neural newtworks aren't the best choice.
There are also good free alternatives like Octave (use 4.0.3 GUI version) or R which both have good NN tools.
Thank you all. Your suggestions were very helpful. I have also attempted to use Statistica for artificial network predictions and it proved a lot better than SPSS, however I got stuck at the point of optimization of the chosen model. I don't know if anyone has used Statistica for artificial neural network modeling. Please help.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software