Frequently asked questions
By Roberto Lopez and Ismael Santana Artelnics.
- What is Neural Designer?
- Who should use Neural Designer?
- What are the main advantages of Neural Designer?
- Is there an special pricing for Universities?
- What project type should I use?
- What data files does the software support?
- Which are the most important tasks?
- How do I split a data set into training, generalization and testing subsets?
- How to test pattern recognition applications?
- How do I make predictions with a trained model?
- Why do I not get the same result when performing twice the same training?
- Is Neural Designer compatible with 32-Bit operating systems?
- Which Linux distributions are supported by Neural Designer?
- What do I do if the proxy server configuration appears?
- Why I cannot import my data file?
- My project file doesn't load - why?
- How do I get support?
Available for Windows, Mac, Ubuntu and Linux Mint; Neural Designer is a very powerful application for predictive analytics and data mining. The software uses neural networks, which is considered the best technique for making predictions. Neural networks are based on using mathematical models of the brain for building complex data representations.
You can use Neural Designer for discovering complex relationships, recognizing unknown patterns, predicting actual trends, recognizing associations from data and much more. The input here is a data set, and the output is a predictive model.
Neural Designer follows an open core model by using the popular open neural networks library OpenNN inside its learning engine.
Neural Designer is a data mining application intended for professional data scientists.
Our most relevant customers are (i) analytics departments at innovative companies, (ii) big data consulting firms and (iii) research centres and university departments.
The users of Neural Designer are specialized technicians, but they work in diverse industries. Some examples are business intelligence, health care or industry 4.0.
With Neural Designer you don't need to write programming code or build complicated block diagrams. The workflow is clearly designed to build and deploy the most powerful predictive models.
Moreover, the software is capable of analyzing bigger data sets in less time, providing our customers with results in a way previously unachievable.
Yes, we offer some special discounts for Universities interested in AI research and data analysis using Neural Designer. You can contact our sales department with this email. They will inform you of our academic licenses and its price.
Neural Designer supports the most common file formats (CSV, Excel, OpenOfficeCalc, Weka...) and databases (Oracle, MySQL, SQLite, SQL Server, Access...).
There are two phases in the development of a neural network: Design and deployment.
- Design phase: Here we use Neural Designer to build a predictive model that learns from data. The most important task in this phase is "Perform training"
- Deployment phase: Here we use our own system to make predictions on new data. The most important task in this phase is "Export expression".
When you load a data file, the uses of the instances are divided by default into training (60%), generalization (20%) and testing (20%) subsets.
In order to change that percentages, go to Task Manager > Data set > Split instances.
In the confusion matrix the rows represent the target classes and the columns the output classes for a testing target data set. The diagonal cells in each table show the number of cases that were correctly classified, and the off-diagonal cells show the misclassified cases.
For the case of two classes the confusion matrix takes the form of the following figure. The green instances are those well classified, while the red instances are those misclassified.
Moreover, for binary classification, there are a set of standard parameters for testing the performance of a neural network as classifier:
- Classification accuracy: Ratio of instances correctly classified. An accuracy of 100% means that the measured values are exactly the same as the given values.
- Error rate: Ratio of instances misclassified
- Sensitivity, or true positive rate: Proportion of actual positive which are predicted positive. For example, a sensitivity of 100% means that the test recognizes all actual positives, i.e., all sick people are recognized as being ill.
- Specifity, or true negative rate: Proportion of actual negative which are predicted negative,.
- Positive likelihood: Likelihood that a predicted positive is an actual positive.
- Negative likelihood: Likelihood that a predicted negative is an actual negative.
For multiple classification, the confusion matrix will appear as in the next figure. The green instances are those well classified, while the red instances are those misclassified.
There are two ways of doing that:
In the Task manager, run Neural network -> Calculate outputs. The following dialog will appear.
In the Task manager, run Neural network -> Write expression. A similar formula to the next one will be written in the viewer.
You can then export that mathematical expression to any programming language (C++, Java, R, etc.).
Before starting the training process, the parameters are perturbed at random for numerical reasons.
The final results depend on many things such as the stopping criteria, the accuracy, etc.
One of them is the initialization of parameters.
However, if the application is well posed you should always obtain similar results.
The first step is to identify the missing values in your data file, and assign them a label. Typical labels used for representing missing values are "NaN" or "?". Do not use numeric values here, such as -999.
When importing the data file, you need to edit the Missing values label in the "Set file properties" page of the "Impor data file dialog".
When the data file is loaded, the software will automatically recognize the missing values and write the corresponding information in the Missing values section of the Data set page.
You can apply two scrubbing methods for missing values. The Unuse method is the default, and makes the instances with missing values to be excluded in the analysis. If your data set is small, you probably cannot afford that, and the Mean method is recommended. This scrubbing method substitutes the missing values with the mean value of the corresponding variable.
An outlier is an observation point that is distant from other observations. They may be due to variability in the measurement or may indicate experimental errors. If possible, outliers should be excluded from the data set. However, detecting that anomalous instances might be very difficult, and usually requires lots of work.
The first thing we can do is to check for the correctness of the data statistics. Indeed, spurious minimums and maximums are a clear sing of the presence of outliers.
We can also plot the data histograms and check that there are not isolated bins at the ends.
Box plots are also a good method for detecting the presence of outliers, since they depict groups of data through their quartiles.
Neural Designer includes two different tasks for cleaning outliers:
- Clean univariate outliers: This task uses the Tukey's method, which defines an outlier as those values of the data set that fall to far from the central point, the median. The maximum distance to the center of the data that is going to be allowed is defined by the cleaning parameter. As it grows, the test becomes less sensitive to outliers but if it is too small, a lot of values will be detected as outliers.
- Clean multivariate outliers: Here a neural network model is built with the entire dataset. Then the maximal errors are calculated. Those points with an error greater than a given value are considered outliers and set to Unused.
Neural Designer also includes the Minkowski error, which is an error term more insensitive to outliers than the sum squared error. The defaul value for the Minkowski parameter is 1.5. If you suspect that there are outliers introducing big errors in your model, you can decrease that value (for instance to 1.25). If you think that the outliers are introducing only small errors, you can increase the Minkowski parameter (for instance to 1.75).
Sometimes the download process can fail. Here are some tips you could check in order to accomplish the downloading.
- On Internet Explorer, the download dialog should appear centered at the bottom and must agree to start the download.
- Firefox should show a dialog box in the center of the page, accepting the dialogue should begin downloading. You can check the download process at the right top.
- In google Chrome the download starts directly and is displayed in the lower left corner.
- On Safari it appears in the upper right corner.
These dialogs may take a few seconds to appear.
In some browsers, once the download is complete, you have to accept a new dialogue to save the installer to your hard drive. This is done for security reasons. The browsers warn about downloading executable files from the internet. Our installer is an "exe" file, for this reason, you could get a warning.
- If you have installed an Adblock in your browser, please turn off and check again.
- Check the antivirus configuration or disable it temporarily.
Return to download page.
The latest version of Neural Designer is developed for 64-Bit architectures exclusively. This is meant to accelerate the processing of big data sets, and offer a better support for the growing number of 64-Bit PC users.
Neural Designer is developed and tested to work in Ubuntu 16.04 LTS and Ubuntu 14.04 LTS. It also works in Linux Mint as it shares a common kernel with Ubuntu. Currently we provide no support for Fedora, Debian and other non-Ubuntu Linux distributions.
Neural Designer is configured to automatically detect a proxy server. If you are in a network under a proxy server, the program will display the configuration fields
so that you can connect to the internet through it. In this case, you must complete at least the "Host name" and "Port" fields as they are mandatory.
The general configuration for connecting to a proxy server on a Windows client machine is managed through the Internet Explorer browser.
Internet ExplorerGo to the top right wheel and select "Internet Options".
Figure 1. Internet Explorer menu.
In the next window, choose the "Connections" tab and then click the "LAN Settings" button.
Figure 2. Internet options.
If you are using a proxy, the "Use a proxy server ..." box in the "Proxy Server" section must be checked and the address and port of the proxy server are displayed. The address is the "Host name". It can come as a domain name or as an IP address.
Figure 3. Internet Explorer LAN Settings.
Try connecting with this information. If this is still not possible, read the following.
Normally, the network administrator will have created users to connect to this proxy server. In this case, you should contact your network administrator to get this information. This configuration is done on the server, so you will not have access to it on your machine. Your network administrator must provide you with an active user and password on the proxy server.
Google ChromeGoogle Chrome uses the default system settings that, as stated above, is set in the Internet Explorer browser. To see this setting if you are using Google Chrome, go to "Settings" by displaying the three vertical dots menu at the top right.
Figure 4. Google Chrome menu.
Once there, scroll to the bottom and click on the "Show advanced settings" link. Go to the "Network" section and click the "Change proxy settings" button.
Figure 5. Google Chrome proxy configuration.
This opens the Internet Explorer dialog box we saw earlier.
Mozilla FirefoxGo to the top menu on the right.
Figure 6. Mozilla Firefox menu.
Click on the "Options" button and then go to the "Advanced" section and click on the "Network" tab.
Figure 7. Mozilla Firefox network configuration
Finally, click the "Settings" button. You will see the following window.
Figure 8. Mozilla Firefox proxy configuration.
Importing a data file is a very delicate process, since it can come in many different formats and might have any content.
If you are having troubles when importing your data file, check the following list:
- The data is arranged in rows (instances) and columns (variables).
- All rows have the same number of columns.
- The file properties (columns name, rows label, separator, missing values label...) have been set properly.
- Columns contain only numerical or nominal values (except for the header line).
- The number of nominal values in a column is not very big (each name is transformed into a numerical variable).
If a NDP file is failing to load it is very suitable that the relative locations between the project file (ndo) and the data file (txt, dat, csv, ...) has changed. The easiest way is to restore that two files to their original locations.
To learn more about Neural Designer, you can look at the tutorials that come with the software. If you have a question, contact the technical support team. A course on neural networks and Neural Designer is your best learning resource.