Polymatica ML

The main articles are:

• Machine Learning
• Data mining Data mining
• Business Intelligence Definition

The Polymatica ML software product is included in the register of domestic software.

2021

Polymatica ML 1.0 Data Mining Module Release

On December 24, 2021, Polymatica, a Russian developer of solutions for analyzing large amounts of data, presented the first commercial version of the Polymatica ML module, which allows business users to independently create machine learning models and control their life cycle. Software development was carried out with a grant from the Russian Information Technology Development Fund (RFRIT).

The first version of Polymatica ML, a data mining module using machine learning methods, allows you to go through the entire cycle of creating and applying models: conduct preliminary data analysis to identify the most significant indicators for building future models; build multiple models using a simple constructor, compare their quality, and define a champion model. In the next step, the user chooses how to apply the model to the main data array - several of them depending on the task being solved. For effective industrial operation, there is a built-in mechanism for monitoring the model, since over time its accuracy decreases and adjustment is required.

List of functions of Polymatica ML 1.0

Data study:

• Generation and visualization of a sample from relational databases.
• Evaluation of sample characteristics according to the main statistical criteria.
• Selection and transformation of characteristics.

Building models:

• Create and configure machine learning models in the GUI.
• Testing and evaluation of model quality.
• Configuring model hyperparameters.
• Model interpretation.
• Connecting Python libraries with machine learning algorithms.

Model Management:

• Centralized model repository.
• Model lifecycle support.
• Export/Import Models.
• Publishing models: relational databases, service (REST API), Polymatica Analytics.
• Compare models from the repository and choose the best model.

Starting to develop the module, we chose two goals: the democratization of machine learning and the creation of a life cycle management platform for machine learning models in Russia. " In a short time, we were able to assemble a team with experience in international companies and create a product that is a designer that allows you to assemble machine learning models from ready-made components, which significantly reduces the time from the formulation of the hypothesis to its verification. At the same time, we went beyond just creating models, targeting the niche of ModelOps or managing machine learning models at the enterprise level. This allows you to store all models in a single repository, systematically monitor the quality of their work and reuse existing algorithms in new models. In the first version, this functionality is already available, and we plan to actively develop it in the near future, " commented Konstantin Malashenko, Product Director, Polymatica ML, Polymatica

The practical implementation of Polymatica ML is already gaining momentum: the module is involved in several pilot projects in different industries: predicting the response to advertising campaigns for retail, predicting the breakdown of heat supply pipes for power, monitoring the state of health in harmful industries.

Prototype presentation

On September 3, 2021, Polymatica presented a prototype of the Polymatica ML product, which will help users investigate data and create machine learning models without resorting to programming, quickly test various hypotheses for their use, and then implement them into the organization's business processes. The module also provides opportunities for managing the lifecycle of ML models, which makes it special in the Russian IT market.

In 2020, the company presented a project to develop its own data mining module using machine learning methods - Polymatica ML.

The Polymatica ML module consists of three functional components. Using Data Discovery, the user connects to various data sources and generates samples for analysis. During the study, he can identify relationships and patterns in the data structure, determine the presence of omissions and anomalies, and test hypotheses. The tools for profiling, visualization, correlation matrices and cluster analysis developed by Polymatica specialists allow you to select, among other things, the most important variables for further modeling.

The second component, Model Designer, is designed to build machine learning models. The visual designer allows you to create models using all the current algorithms: from the decision tree to gradient boosting, ensembles and neural networks. Also, the user can assess the quality of models, compare them with each other according to the selected metrics and choose the best.

The Model Manager component is a centralized repository where you can add models created using Polymatica ML and import them from other analytical tools. The mechanism developed by Polymatica allows you to control the life cycle of the model, including by customizing each of its stages. The user can also retrain the model on new data if its accuracy does not meet the current requirements, and then publish it again. Model Manager helps to build transparent work with each machine learning model in the repository so that it brings maximum business benefits.

Models developed with Polymatica ML can be used by various divisions of the organization - the developers have provided flexible and understandable tools for managing security and access rights. The module can also be modified if necessary, if the company has special requirements.

When creating Polymatica ML, we analyzed the best world experience, attracted to our development team and consultants with the highest level of knowledge about machine learning and product development experience. We set ourselves the goal of developing a convenient and effective tool aimed at simplifying the solution of government and business problems related to advanced analytics and machine learning. I think that we succeeded. By September 2021, the development of the MVP version product has been completed, on the basis of which we will soon conduct a number of pilots for our customers. The release of the commercial version is scheduled for the end of 2021, - said Konstantin Malashenko, product director of Polymatica ML, Polymatika.