 |
Technology, Computation and Data Analysis |
JTerm - Linux commands for Windows?
February 21, 2018 Reading time: ~1 minute
JTerm is an attempt to mimic Linux/Unix commands using 100% Java. Thus it works on Windows. This project is its infancy, but it is already very useful. For some, it may also look also more attractive than using Cygwin on Windows. First thing to notice - it is only 3M in size, a tiny program compared to the default Cygwin installation. Secondly, it has nice look and feel thanks to Nimbus L&F. In combination with JPort portable Java desktop, it may lead to interesting alternative to Cygwin.
The Mutilation of Wikipedia
February 21, 2018 Reading time: 7 minutes
There was a time when I just loved Wikipedia. In the early 1980s, the editors of the Oxford English Dictionary (OED) were planning a second edition of their semi-historical document but they realized they could not possibly handle all the cross-referencing and indexing of the envisaged new edition. The solution turned to be a major project consisting in "computerizing" the OED. They succeeded in getting successive editions but only after much effort.
DataMelt article on Wkipedia was vandalized
January 21, 2018 Reading time: ~1 minute
After 10 years in Wikipedia, a few editors have started a campaign to remove DataMelt Wikipedia article. Impossible requirements may hit 90% of Wikipedia content on free programs in Numerical software and Free science software (and more!) categories.
Using Neural Networks to create titles for scientific papers
February 22, 2017 Reading time: ~1 minute
Neural networks are getting smart and their outputs become increasingly realistic. Here is a neural network example created by Jahred Adelman who used the Multi-Layer Recurrent Neural Network (RNN package) designed for character-level language models. The neural net was trained using 600,000+ real paper titles taken from INSPIRE. Then it was used to create "fake" paper titles. Use this link to find which article title is fake and which is real.
S.Chekanov (ANL)
Top 40 Statistical Data Analysis Software Programs
February 21, 2017 Reading time: 12 minutes
Here is a list of 40 most popular commercial and free statistical and data analysis programs. The emphasis of this software listing is on financial and marketing applications. For more general and technical information please refer the corresponding links on the original source.
DataMelt
February 21, 2017 Reading time: 2 minutes
DataMelt ("DMelt") is a free software for numeric computation, mathematics, statistics, symbolic calculations, data analysis and data visualization. This program combines the simplicity of scripting languages, such as Python, Ruby, Groovy (and others), with hundreds of Java numerical and graphical packages. Unlike other similar programs, DataMelt can be used for scientific computations using Java, word"s most-popular enterprise programming language. The program runs on Windows/Mac OS/Linux and Android platform. DataMelt is a modern version of jHepWork (2005-2013) and SCaVis (2013-2015) programs.
RTextDoc. Editor for LaTeX / AsciiDoc structured documents
February 21, 2017 Reading time: ~1 minute
RTextDoc, editor written Java, is getting rather mature. Why another editor? RTextDoc is the only portable editor that runs from USB on any platform (Windows, Linux, Mac OS). It is the only editor that has instant grammar checking for 15 languages - a feature which is missing for all available LaTeX editors. Finally, it is the only editor for AsciiDoc documents
Recasting Java neural networks in Python
February 22, 2016 Reading time: 2 minutes
Many neural network applications implemented in Java, such as Neuroph, Encog and Joone, may look rather different when switching from the Java language to Python with the help of the DMelt computing environment. First of all, they look simpler. You can use your favorite Python tricks to load and display data. The Python coding is simpler for viewing and fast modifications. It does not require recompiling after each change. At the same time, the platform independence and multi-threading environment are guaranteed by the Java Virtual Machine.
As a simple example, consider Joone, the Java Object Oriented Neural Network. It is well-tested open source project that offers a highly adaptable neural networks for Java programmers. For the purposes of this blog, we will teach Joone to recognize a very simple pattern, such as XOR. The XOR operation's truth table is summarized as
[[0.0, 0.0, 0.0],[0.0, 1.0, 1.0],[1.0, 0.0, 1.0],[1.0, 1.0, 0.0]]
where three first two columns are inputs, and the last column is the output. Now let us put together a simple Python code that uses Joone's Java API:
http://datamelt.org/code/code.php?id=49760300.py
We assume that the neural network uses two input layers, one hidden (with 3 neurons), and one output layer (with one neuron). Now, start the DMelt program, and run this code in its editor. If you are using Mac/Linux, you can run this code in a batch mode with the command "dmelt_batch.sh code.py". You will see how the global error changes as a function of epoch. This trend is shown here.
Joone can be used for rather sophisticated tasks that involve future predictions. A popular example using Python calling Joone library deals with forecast in time series. For this task, we generate a time series in the form cos(x)*sin(x), add nose, and use this as input for our neural network training. The output image of our forecast code implemented in Python is shown here.
See more neural network examples implemented in Python/Jython and Java in DMelt Neural Network code examples.