Neural Designer's technical features
Some of the main algorithms it contains are listed below.
Application types
- Approximation (or modeling) to discover intricate relationships.
- Classification (or pattern recognition) to recognize complex patterns.
- Forecasting (or time series prediction) to predict trends.
Data set
- Compatible with the most common data files: CSV, DAT, TXT, Excel, OpenOffice …
- Complete configuration of variables.
- Complete configuration of instances.
- Exhaustive descriptive statistics.
- Estimation of variables importance using many types of correlations.
- Innovative utilities for outlier detection and data filtering.
Neural network
- Network architecture with an unlimited number of layers.
- Perceptron layer with several activation functions: linear, hyperbolic tangent, logistic, and rectified linear.
- Probabilistic layer with different activation functions: binary, continuous, competitive, and softmax.
- Scaling layer with minimum/maximum and mean/standard deviation methods.
- Unscaling layer with minimum/maximum and mean/standard deviation methods.
Training strategy
- Loss indices for all types of applications:
- Weighted squared error for common data sets.
- Minkowski error for dealing with outliers.
- Cross-entropy error for pattern recognition.
- Weighted squared error for unbalanced data sets.
- L1 and L2 Regularization for avoiding overfitting.
- Optimization algorithms for all volumes of data:
- Stochastic gradient descent and adaptive moment estimation for huge data sets.
- Gradient descent, conjugate gradient
and adaptative linear momentum for the training of big data sets. - Quasi-Newton method for fast training of medium data sets.
- Levenberg-Marquardt algorithm for high-speed training of small data sets.
Model selection
- Neurons selection algorithm for finding the optimal network architecture: incremental order.
- Inputs selection algorithms for selecting the most important features: growing inputs, pruning inputs, and genetic algorithm.
Testing analysis
- Testing errors and corresponding
statistics and
histogram
calculation. - Linear regression analysis for functional regression problems.
- Confusion matrix for pattern recognition applications.
- Full set of metrics for the evaluation of binary classification tests.
- ROC curve for diagnostic tests.
- List of misclassified instances.
- Cumulative gain and lift charts for segmentation applications in marketing.
- Profit chart for simulating marketing campaigns.
Model deployment
- Calculation of output values.
- Directional plots for exploring the predictive model.
- Exportable mathematical expression of the model.
- Exportable predictive model in C and Python.
Output
- Exhaustive results in interactive report plenty of descriptions, tables, and figures.
- Report exportable to Word and Pdf formats.
- Results exportable in data format.
Help
- Simple user's guide.
- Extensive tutorials on the application of neural networks with Neural Designer.
- Several step-by-step solved examples of machine learning applications in different fields.
- Premium technical support by email, phone, or video chat.
Performance
- Software developed with the high-performance programing language C++.
- Code subjected to optimization techniques for memory management and processing speed.
- CPU parallelization employing OpenMP.
- GPU acceleration with CUDA, MKL and Eigen.
Supported platforms
- Windows.
- Mac OS X.
- Linux (Debian and Ubuntu).