Scientific Consultant Services, Inc.
N-TRAIN® NEURAL NETWORKS
N-TRAIN®
Neural Network Development Tools
Neural Network Applications
Neural networks are employed for an ever increasing variety of
purposes, including (but not limited to):
- Human gait identification
- Facial recognition in crowded fields
- Automatic clutter elimination and target identification in
radar
- SIGINT (signals intelligence)
- Cognitive radio, software-defined radio, and weak signal
enhancement
- Semantic analysis and classification of text (e.g.,
classification of web pages)
- Voice recognition with automatic transcription to text
- Optical character recognition (including handwriting)
- Satellite image enhancement and analysis (including
battlefield analysis)
- Automatic tumor detection in MRI and CAT scans
- Automatic EKG analysis and interpretation
- Galaxy classification in astronomy
- Meteorological forecasting
- Chemometrics and spectroscopy (including forensic
applications)
- Unmanned aerial vehicle (UAE, drone) flight control and
stabilization systems
- Automated sawmill operation
- Industrial control systems
- Insurance risk prediction
- Direct mail response prediction
- Options pricing and futures trading
- Seismological and earth sounding data reduction
- General pattern recognition and classification
Design Considerations and Performance
N-TRAIN, which may be used for any of the above, is a
neural network development system designed specifically for speed,
capacity and numeric stability. Double-precision
calculations carried out in hand-optimized assembly provide N-TRAIN
with exceptional numerical stability. With N-TRAIN, neural
networks can be trained reliably and quickly, even with
"difficult" data sets and when other packages fail to
converge. N-TRAIN also performs remarkably on noisy data
(such as that from the financial markets and weak signal detection
systems) as well as on extremely large data sets.
Several training models, including standard backpropagation and
genetic evolution (which is more resistant to local optima), may
be specified. Sophisticated analysis of the significance of
each input variable to a model is easily performed. Finally,
complex training and testing procedures may be automated with
scripts. N-TRAIN is designed to run optimally on Linux and
Unix systems (including the Mac); however, it will run under
Cygwin in Windows.
The product has had great reviews in Technical Analysis of
Stocks & Commodities and in Futures (a
comparative review by Mark Jurik). It was employed in The Encyclopedia of Trading
Strategies (Jeffrey Owen Katz & Donna McCormick;
McGraw Hill, 2000), to develop a "universal forecaster"; in Computerized
Trading (Ed. Mark Jurik; Prentice Hall, 1999) for pattern
recognition; and, in Advanced
Options Pricing Models (Jeffrey
Owen Katz & Donna McCormick; McGraw Hill, 2005) for
pricing options. It has been used by several
corporations and numerous governmental agencies. Many users
have said that N-TRAIN is faster, provides better
generalisation, and is more robust than other packages they have
tried.
In his review, John Sweeney (former Editor of Technical
Analysis of Stocks and Commodities) said the following
about N-TRAIN:
This is the real thing. With N-Train there
is no waiting. The program is very, very fast.
N-Train converged quickly in all the examples I gave it.
In "Working With Neural Networks," published in Club 3000
News, Ray Wertheim said:
I have either used or tried four other
commercially available products. The product I have
found most useful for my work is called N-Train. This
software allows you to write data directly into files that can
then be scaled and trained. The trained network can then
be utilized in a trading system from within TradeStation.
Most impressive of all was an independent, comparative review of
about a dozen neural network development packages published by
Mark Jurik in Futures Magazine ("Consumer's Guide to
Neural Network Software"). When the reviewer's scores were
ranked for speed, test bias (generalization), and variance
(consistency), and the ranks then averaged, N-TRAIN ranked
best in overall performance!
Features
- No inherent limit to problem size other than memory and
hard-disk space
- High degree of user control: global and layer-specific
transfer functions, error measurements, and learning rates,
among other things, may be set by the user
- Built-in scripting language for complete automation of
training and testing; consistent design means that anything
done in interactive mode can be done in scripts
- Checkpointing (the network's state is saved at intervals
during training)
- Thorough fact shuffling
- Data consistency and validity checking
- May be used as a neural network engine from within other
applications and programming languages (e.g., Fortran, C++,
Python, Bash)
- Uses standard file formats easily read and written by other
software packages (e.g., Octave or Matlab, Open Office, all
standard unix tools, popular programming languages)
- Libraries included: N-Train's functionality may be directly
embedded in other applications
- Multi-platform: N-TRAIN and its libraries will run, for
example, on a Raspberry Pi
- Detailed manual in .pdf format
- Genetic training (breeding populations of neural networks)
is supported
- Regression models are supported
- Partially connected neural networks are supported
- Probabilistic and statistical classifier networks are
supported
- Variable relevance analysis (by best-regression variable
replacement estimation)
- Statistical analysis with histograms, deciles,
correlations, and more
- Correlation module to compute standard Pearson
Product-Moment correlations
- Variable orthogonalization using SVD (to handle excessive
co-linearity) is supported
- Automatic and manual scaling methods are supported
- Multiple vote models employing a set of neural networks are
supported
Package Contents and Purchase
N-TRAIN comes with all source code. Also bundled with the
product is "N-TRAIN for C", the N-TRAIN neural network library for
embedded applications. N-TRAIN may be purchased from the software page using PayPal; the
software is delivered as a zipfile sent to the purchaser's email
address. The total cost is $495.00 which includes basic
technical support.
Consulting
Consulting on neural networks and their applications in diverse
fields (as well as on N-TRAIN itself) is available. Inquire
by email to: jeffkatz@scientific-consultants.com
Copyright © 2013 Scientific
Consultants Services, Inc. All Rights Reserved.
