Summary
Neural Designer is a software tool for advanced analytics, and includes tools for descriptive, diagnostic, predictive and prescriptive analytics. The objective is to get actionable insights resulting in smarter decisions and better business outcomes.
The software is based on neural networks, which are considered one of the most powerful technique for data analysis. One of the advantages of Neural Designer is a graphical user interface that clearly defines the workflow and provides comprehensive results.
Graphical user interface
The graphical user interface of Neural Designer consists of a sequence of steps that simplify data entry. There is no need for programming or building block diagrams.
The software comes with tutorials and examples for users to learn how to use it from scratch.
The results from Neural Designer are written in a report displaying comprehensive insights from all tasks. Users can visualize the results with tables, charts and pictures that can be exported to other tools.
Also, the whole report displayed in the viewer can be exported to ODT and PDF formats.
Analytics algorithms
Neural Designer contains a large range of advanced algorithms that allow data scientists to build powerful models. The following list summarizes the algorithms included in the software.
Data Set
- Descriptive statistics.
- Variables histograms.
- Box plots.
- Scatter charts.
- Inputs correlation matrix.
- Inputs-targets linear and logistic correlations.
- Advanced methods for data balancing.
- Innovative utilities for outlier detection.
- Principal components analysis.
Neural Network
- Network architecture with unlimited number of layers.
- Threshold, symmetric threshold, logistic, hyperbolic tangent and linear activation functions.
- Scaling and unscaling layers with minimum/maximum and mean/standard deviation methods.
- Probabilistic layer with binary and softmax methods.
Loss Index
- Sum squared error, mean squared error, root mean squared error and normalized squared error functionals for common data sets.
- Minkowski error for dealing with outliers.
- Cross-entropy error for pattern recognition.
- Weighted squared error for unbalanced data sets.
- Regularization for avoiding overfitting.
Training Strategy
- Gradient descent and conjugate gradient for training of big data sets.
- Quasi-Newton method for fast training of medium data sets.
- Levenberg-Marquardt algorithm for very fast training of small data sets.
Model Selection
- Incremental order and simulated annealing for finding the optimal network architecture.
- Growing inputs, pruning inputs and genetic algorithm for selecting the most important features.
- Threshold selection though F1 score, Matthew’s correlation, Youden’s index, Kappa coefficient and ROC curve distance optimization.
Testing Analysis
- Complete error data and corresponding statistics calculation.
- Linear regression analysis for function regression problems.
- Confusion matrix for pattern recognition applications.
- Full set of metrics for evaluation of binary classifiers.
- ROC curve for diagnostic tests.
- Cumulative gain and lift charts for segmentation applications in marketing.
- Calibration plot for classification problems.
- Error autocorrelation and cross-correlation for time series prediction.
- List of misclassified instances.
Model deployment
Neural Designer provides an easy way for deploying predictive models. For that, you can use a standard such as PMML, or export the resulting model to programming languages such as R or Python.
An example of that expression in the R programming language is written below.
expression <- function(x) {
scaled_x<-2*(x+1)/(1+1)-1
y_1_1<-tanh(0.361707-0.497807*scaled_x)
y_1_2<-tanh(-0.15776-0.376231*scaled_x)
y_1_3<-tanh(0.295148+0.493422*scaled_x)
scaled_variable_2<-(-0.0531044-0.868102*y_1_1-0.778027*y_1_2+0.818376*y_1_3)
outputs <- c(0.5*(scaled_variable_2+1.0)*(1+1)-1)
outputs
}
Availability
Neural Designer is a proprietary software. However, there is a 15 day free trial available to download from www.neuraldesigner.com.