{"id":41221,"date":"2020-02-06T20:42:46","date_gmt":"2020-02-06T17:42:46","guid":{"rendered":"https:\/\/prohoster.info\/blog\/blog_prohoster\/python-gateway-v-intersystems-iris"},"modified":"2020-02-06T20:42:46","modified_gmt":"2020-02-06T17:42:46","slug":"python-gateway-v-intersystems-iris","status":"publish","type":"post","link":"https:\/\/prohoster.info\/en\/blog\/python-gateway-v-intersystems-iris","title":{"rendered":"Python Gateway in InterSystems IRIS","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

This article is dedicated to Python Gateway \u2014 a community project with open-source code for the InterSystems IRIS data platform. This project allows orchestration of any machine learning algorithms developed in Python (the primary environment for many Data Scientists), and utilize numerous ready-made libraries for quickly creating adaptive, robotic analytical AI\/ML solutions on the InterSystems IRIS platform. In this article, I will show how InterSystems IRIS can orchestrate processes in Python, effectively manage bidirectional data transfer, and create intelligent business processes.<\/p>\n

<\/a><\/noindex><\/p>\n

Plan<\/h1>\n

<\/p>\n

    \n
  1. Introduction.<\/li>\n
  2. Toolkit.<\/li>\n
  3. Installation.<\/li>\n
  4. API.<\/li>\n
  5. Interoperability.<\/li>\n
  6. Jupyter Notebook.<\/li>\n
  7. Conclusions.<\/li>\n
  8. Links.<\/li>\n
  9. MLToolkit.<\/li>\n<\/ol>\n

    <\/p>\n

    Introduction<\/h1>\n

    <\/p>\n

    Python is a high-level general-purpose programming language aimed at increasing developer productivity and code readability. In this series of articles, I will discuss the capabilities of Python on the InterSystems IRIS platform, with the primary focus of this article being the application of Python as a language for creating and applying machine learning models.<\/p>\n

    <\/p>\n

    Machine learning (ML) is a class of artificial intelligence methods characterized by not directly solving a problem, but rather learning from solving many similar problems. <\/p>\n

    <\/p>\n

    Machine learning algorithms and models are becoming increasingly widespread. There are many reasons for this, but it all boils down to availability, simplicity, and the achievement of practical results. Is clustering or even neural network modeling a new technology? <\/p>\n

    <\/p>\n

    Of course not, but today there is no need to write hundreds of thousands of lines of code to run a model, and the costs associated with creating and applying models are continually decreasing.<\/p>\n

    <\/p>\n

    Tools are evolving \u2014 while we do not have fully GUI-oriented AI\/ML tools, the progress we have seen with many other classes of information systems, such as BI (from coding to using frameworks and GUI-oriented configurable solutions), is also observed in tools for creating AI\/ML. We have already moved beyond the coding stage and now utilize frameworks for building and training models.<\/p>\n

    <\/p>\n

    Other improvements, such as the ability to deploy pre-trained models where the end user simply has to complete the training on their specific data, also simplify the initiation of machine learning. These advancements significantly ease the learning curve for machine learning for both specialists and companies as a whole.<\/p>\n

    <\/p>\n

    On the other hand, we are collecting more and more data. Thanks to a unified data platform like InterSystems IRIS, all this information can be immediately prepared and used as input for machine learning models.<\/p>\n

    <\/p>\n

    With the shift to the cloud, launching AI\/ML projects becomes easier than ever. We can consume only the resources we need. Moreover, thanks to the parallelization offered by cloud platforms, we can save on time spent.<\/p>\n

    <\/p>\n

    But what about the results? Here it gets more complicated. There are numerous tools for building models, which I will discuss next. Creating a good model is not easy, but then what? Deriving value from the model's use in business is also a non-trivial task. The root of the problem lies in separating analytical and transactional loads and data models. When we train a model, we usually do this on historical data. However, the place for the trained model is in transactional data processing. What good is the best model for detecting fraudulent transactions if we only run it once a day? The fraudsters are long gone with the money. We need to train the model on historical data, but we also have to apply it in real-time to incoming data so that our business processes can operate according to the predictions made by the model.<\/p>\n

    <\/p>\n

    The ML Toolkit is a set of tools designed specifically for this purpose: to merge models and transactional environments so that the built models can be easily utilized directly in your business processes. The Python Gateway is part of the ML Toolkit and provides integration with the Python language (similarly, R Gateway, which is also part of the ML Toolkit, provides integration with the R language).<\/p>\n

    <\/p>\n

    Toolset<\/h1>\n

    <\/p>\n

    Before we continue, I would like to describe a few tools and libraries for Python that we will be using later.<\/p>\n

    <\/p>\n

    Technologies<\/h2>\n

    <\/p>\n