CMSR Data Miner / Machine Learning / Rule Engine Studio
Predictive Modeling by Machine Learning & Rule Engine & Data Mining

CMSR Data Miner / Machine Learning / Rule Engine Studio (CMSR Studio for short) provides an integrated environment for machine learning predictive modeling, expert system shell rule engine, and big-data data mining. CMSR is a perfect platform to develop advanced predictive models using deep learning techniques, for business data, combining predictive models and business rules. Deploy models on the web and Android devices, and embed machine learning predictive models into business applications.

See CMSR Studio Screenshots.

(Full Model Development Life Cycle Support) Data Analysis, Variable Relevancy/Principal Factor Analysis, (Rule-based) Data Transformation, Modeling, Model Integration, Deep Learning with Rule Engine, Model Validation & Fitness Testing, Database Scoring & Updates, Model Deployment on the web and Android and embedded applications. CMSR is a Swiss Army Knife of machine learning, predictive modeling and data science. It covers full model development life cycle.

(Graphical User Interface and Powerful Visualization Tools) CMSR provides easy-to-use Graphical User Interfaces (GUI) with intuitive powerful model and data visualization tools. You can develop advanced powerful models very easily with GUI interface. No need for programming long cranky codes.

Very powerful graphical visualizations, advanced power options and smooth integration with database systems make CMSR well-suited for advanced power users as well as beginners.

(Optimize for Model Accuracy and Generality) The state art of predictive modeling is the development of models that are both accurate and general at the same time. These conflicting goals are difficult to achieve without advanced tools. It requires more than API calls or web browser interface. Step-by-step continuous incremental training with multiple datasets. Network fine tuning. Network pruning and compression. Testing and validation. Database scoring. Model characteristics/behavior analysis. These require advanced interactive graphical analytic visualization tools that combine database (update), data visualization and special tools for boutique power model developers. All these can be done using CMSR Studio.

(Fast Enterprise Performance) CMSR employs fast in-memory technology. This makes CMSR very fast. Most operations finish within a second on million records. Most database visualization and reporting tools employ SQL GROUP-BY operations. Simple to deploy advanced sophisticated models on Rosella BI Web Server and on Android devices. Very fast! Rosella BI Server can process complex RME-EP deep learning model HTTP/JSON requests over 5,000 times per second on average dual core computers. (Note that performance can vary depending on neural network sizes.)

(Model Code Generation for Embedded Applications) Designed for developing AI/ML-based business & medical solutions, and embedded applications. Predictive models (neural network, SOM, decision tree, regression) can generate (program function) source code in Python, PHP, Java/JSP/Android, C/C++/Objective-C, C#, VB, Excel/VBA, Swift, Ruby, JavaScript/Node.js. CNN and OD-CNN can generate source codes in Java, C/C++, Swift(CNN only). It is self contained program function code and multi-thread safe. You don't need to hassle with complex libraries, DLLs, and dependencies! It's really fast and compact. Small memory requirement. Just paste into your programs and add function call statements. No need to convert data into tensors or arrays. Perfect for developing Android, iOS and web applications.

Excellent algorithms, advanced power tools, super fast, big data (up to 2 billion records), easy-to-use GUI interfaces, smooth database incorporation, superb graphics and user experience!

Applications of Neural Network

• ANN (Artificial Neural Network): Predictive modeling. Risk management for credit, finance, insurance, etc.
• CNN (Convolutional Neural Network): Computer vision. Visit CMSR Powered Skin Cancer Check website.
• OD-CNN (Object Detection CNN): Computer vision. Wildfire/Bushfire detection.
• M-CNN (Multivariable Measuring Convolutional Neural Network): Computer vision. Wildfire/Bushfire detection.
• SOM (Self Organizing Maps): Clustering and (customer) segmentation.

(1) Rule Engine for Machine Learning / Deep Learning / Expert Systems

• RME-EP: Rule Engine with Machine Learning, Deep Learning, Neural Network.
• Easy to build "Deep Learning" hierarchically stacked multiple neural network models.
• Combines rule engine and machine learning for advanced predictive modeling. It's very easy to incorporate various machine learning algorithms into one single RME-EP model using rules, such as neural networks, SOMs, decision trees, ensembles, non-random forests, radial basis functions, regression, time-series, etc.
• Employs forward-chaining rule inference engine.
• Rules can be written to incorporate predictive models.
• Powerful easy-to-learn rule specification language (derived from SQL).
• Easy-to-use Machine Learning Studio.
• Can generate bagging, deep learning, predictive model evaluation rules automatically.
• Can be used as rule-based data transformation tools.
• Implement Semi-supervised Learning, Radial Basis Function (RBF) using SOM and Neural Network models.
• Support regression, time-series and statistical functions.
• Randomization rules can be used to create sampling datasets from database tables.
• Database scoring: Apply RME-EP models to database records and can update multiple fields.
• Models are deployed on Web and Android.

Want to develop super intelligent models? Start with the example in this page and expand it!

(2) Predictive Modeling by Machine Learning

• Neural Network: ANN / MLP Multi-layer Deep Neural Network. Optimized for statistical predictive modeling. Support sigmoid (or logistic), hyperbolic tangent, softmax and softmax+(normalized) activation functions. (Continuous incremental learning with multiple training datasets, network pruning and compression, random data training, best model search, network evolution diagram, GPU JOCL/OpenCL computing support.) Graphical interface and visualization helps to develop highly optimized neural network models. For examples of neural network, please read the following links;
  • Credit Risk Predictive Modeling.
  • Insurance Risk Predictive Modeling.
  • Neural Network Time-series Forecasting with Seasonal Adjustment.
• SOM: Self Organizing Maps for Clustering, Segmentation and Semi-supervised/Unsupervised learning. (Amazingly powerful tool!)
• Decision Tree (Cramer, Entrophy, GINI, Chisquare, Expected Accuracy, etc.) for classification and segmentation. (Note that Cramer tends to produce trees with fewer nodes but more accurate than other well-known decision tree algorithms.) Using multiple trees and RME-EP rule engine, develop non-random forests and ensembles.
• Regression supporting categorical/nominal variables and searches.
• Above predictive models can generate program (function) source code in JavaScript/Node.js, Python, PHP, Java/JSP, C/C++/Objective-C, C#, VB and Excel/VBA, Swift, Ruby.
• In addition to source code integration, models are deployed on Web and Android.
• LSTM : Available for Java APIs.
• CNN (Convolutional Neural Network for Computer Vision, Image Classification, OCR) : Powerful Easy To Use GUI interface! Supports both GPU and CPU core parallelism. Super efficient program source code generation in Java, C/C++, Swift for embedded application development. (Developer license is required for code generation.) Support incremental training with different training parameters and datasets. Excellent visualization of internal model data can help you to understand models better. Support inversion, conversion to gray images, and histogram equalization. Very easy data preparation. Support data augmentation. No need to write complex coding! Powerful advanced features. Well suited for drone and IoT/AIot applications. Big multi-tera bytes in-memory technology. For more information, please read Convolutional Neural Network.
• OD-CNN (Object Detection CNN): For object detection with classification. Supports YOLO-style modeling! Powerful GUI interface. Supports both GPU and CPU core parallelism.. Incremental training. Data labeling software provided. Support data augmentation. Big multi-tera bytes in-memory technology. Super efficient program source code generation in Java and C/C++ (IEEE754 compatible) for embedded application development. (Developer license is required for code generation.) Well suited for drone and IoT/AIot applications. For more information, please read Convolutional Neural Network and Wildfire detection by Machine Learning.
• M-CNN (Multivariable regression or Measuring Convolutional Neural Network for Computer Vision) : Powerful Easy To Use GUI interface! Supports both GPU and CPU core parallelism. Super efficient program source code generation in Java, C/C++, Swift for embedded application development. (Developer license is required for code generation.)
• Supports Rule-based predictive model integration.
• Visualization tools for predictive modeling variable relevancy analysis.
• Database scoring: Apply predictive models to database table records.

(3) Time-series and Date/Time Dimension Analysis Tools

• Time-series reports (with regression, seasonal adjustment, charts).
• Group-by reports (with regression, seasonal adjustment, charts).
• Time-series similarity reports (with charts).
• Neural Network Time-series Forecasting with Seasonal Adjustment.
• Dynamic Active Visualization Charts: 3D bars, bars, lines, histograms, Sankey diagram, histobars, scatterplots, boxplots, and more.
• RME-EP rule engine.

(4) Database Analytics for Big Data: Hive/Hadoop, MySQL/MariaDB, SQL Server, PostgreSQL, ...

• Cross table reports with deviation analysis and highlight.
• Group-by reports with deviation analysis, highlight, charts, time-series/seasonal adjustment.
• Time-series analysis for big data: group-by time-series trend analysis using regression, incorporating smoothing and seasonal adjustment. (In the following figure, green columns are time-series data. Orange columns are projected values from regression with seasonal adjustment.)
  
• Dynamic active power visual charts: 3D bars, bars, pies, and more.
• Database metadata navigation and data browser.
• Data import and export tools.
• SQL query/batch tools.

For more, please read Big Data Analytics.

(5) Clustering and Segmentation: Customer Prioritizing and Targeting

• Neural Clustering (SOM) with Response and Profit Modeling/Analysis.
• Decision Tree with Response and Profit Modeling/Analysis.
• Database scoring: Apply segmentation models to database records.

For step-by-step guides for customer segmentation, prioritizing and targeting,
please read Customer Prioritizing and Segmentation (PDF).

(6) Hotspot Drill-down and Profiling

• Hotspot Drill-down Analysis and Profiling.

(7) Cross-sell Shopping Basket Analysis

• Association Rules - Apriori Algorithm (Works with both database and file data. Big data!)
• Co-items Shopping Basket Analysis.

(8) Data Mining and Statistics

• Cross table reports with deviation analysis and highlight.
• Group-by reports with deviation analysis, highlight, charts, time-series/seasonal adjustment.
• Dynamic active visual charts: 3D bars, bars, lines, histograms, Sankey diagram, histobars, scatterplots, boxplots, and more.

(9) Special Features

• Fast in-memory technology.
• 64-bit integer and floating point data representation.
• Large in-memory optimized for fast speed (up to tera bytes).
• Super fast and Big data (single dataset up to 256 tera-bytes).
• Connect to and update major relational DBMS through JDBC.
• Data Miner optimized for MicroSoft MS SQL Server, MySQL / MariaDB, PostgreSQL, MS Office Access, Apache Hive + Hadoop.
• Deploy/publish predictive models on Android phones and Android tablets with MyDataSay app. For downloads, click here.
• Runs on multiple OS: Windows, Mac OS X / MacOS.

(Note that StarProbe data miner is replaced with CMSR Data Miner / Machine Learning / Rule Engine Studio.)