It has been over a year since I have posted last here. We have learned a substantial amount and finally have a cloud product that is about ready to be sold in the near future. I plan to use this training page of the website to write training articles and provide associated training videos. So let me begin by giving you an overview of SkyRELR.
SkyRELR has taken shape to be essentially what I originally imagined. It is both a GUI and an API controlled through IPython Notebook. The GUI is extremely easy to use and has few options that essentially go to the type of model that you wish to build on the MainGUI page of SkyRELR machine learning app. This includes a parsimonious feature selection model using our Explicit RELR methods which are often more effective once you have roughly 500 target responses, as compared to an Implicit RELR model which is most effective usually with smaller numbers of target responses but has the disadvantage that it also usually selects a large number of predictive features. Both Binary and Ordinal RELR models can be developed through the GUI; Multinomial RELR models have to be developed by calling separate Binary RELR models and this only can be done through the IPython Notebook. We have not yet implemented some of the methods that are described in the book – Calculus of Thought. For example, we have not yet implemented RELR’s Survival Analysis, Causal Learning and Sequential Learning. But we expect to gradually introduce those methods as new scripts in the IPython Notebook over the next year.
The Ipython Notebook is the most powerful aspect of SkyRELR. Through the Notebook, you can embed the RELR API in Python code and have SkyRELR do whatever you can imagine and much beyond what is pre-programmed through the GUI. The Ipython Notebook also gives you access to Linux shell commands; we have an example Notebook of how to access these commands. So through the Notebook, you will understand that SkyRELR is not really just a Python-based web app. Instead, it is a Linux machine and you have the capability to understand much of what is happening in this machine and the Linux operating system through the Notebook interface.
What you get when you purchase a product out here through the Purchase Product page of this website is access to a 2 core 3.75 GiB RAM Compute Optimized C3.large EC2 instance that runs at Amazon Web Services (AWS). Your access period can be two weeks or four weeks at present. The 2 core AWS C3.large EC2 instance runs about 10 times faster than my 2 core personal notebook computer, as even AWS instances with a small number of cores seem very efficient. Yet, the primary applications of this are to get acquainted with SkyRELR for training purposes and for a number of one-off pilot model builds in smaller problems where you do not need the horsepower of a lot of parallel processors and RAM.
For larger problems, we recommend against this introductory product with 2 cores. Instead, we recommend a larger machine. For production environments where you will be building a large number of models on a regular basis, you will also need access to the SkyRELR AMI (Amazon Machine Image). With access to our AMI, you will be able to set up your production environment very efficiently with SkyRELR. The SkyRELR instance can run all by itself, as data can be read and written through the IPython Notebook from and to anywhere with a web address or with an AWS S3 data storage address. Or data can be uploaded and downloaded to and from your SkyRELR instance to your local machine through the simple to use GUIs within SkyRELR. Alternatively, one or more SkyRELR instances can be nodes in a Hadoop cluster where your other nodes are other AWS instances that possibly run PySpark and feed data to the SkyRELR instances for model building and scoring.
That is just an overview of our product. I will be posting much more specific introductory training materials in the near future, so please stay tuned.