1 00:00:07,170 --> 00:00:13,200 So far, we’ve talked about building machine learning models and pipelines. In most practical 2 00:00:13,200 --> 00:00:18,310 applications, the return on investment is obtained when the model or pipeline is put 3 00:00:18,310 --> 00:00:24,060 into production, where it is used to get predictions, or scores, for the new cases. 4 00:00:24,060 --> 00:00:31,039 Let’s look back at our overview of different tool categories. In this unit, Model Deployment 5 00:00:31,039 --> 00:00:32,980 is our focus. 6 00:00:32,980 --> 00:00:37,450 Suppose you worked hard to create the best possible machine learning model and the data 7 00:00:37,450 --> 00:00:41,710 preparation pipeline for it. How will you deploy your models? 8 00:00:41,710 --> 00:00:47,790 In many practical scenarios, models are built and deployed by different teams, using different 9 00:00:47,790 --> 00:00:52,910 programming, and perhaps human languages. The teams will use different computing and 10 00:00:52,910 --> 00:00:58,030 data storage environments, and It might prove difficult to translate your program and the 11 00:00:58,030 --> 00:01:05,140 associated data preparation and post-processing steps from one environment to the other. 12 00:01:05,140 --> 00:01:09,910 Currently there are several approaches you can use to solve this problem, some commercial, 13 00:01:09,910 --> 00:01:16,630 some open source. Yet each one typically supports only a subset of all possible models, from 14 00:01:16,630 --> 00:01:22,630 building them to deploying, so a user gets locked into a specific framework. 15 00:01:22,630 --> 00:01:27,670 Open standards for model deployment are designed to support model exchange between a wider 16 00:01:27,670 --> 00:01:35,350 variety of proprietary and open source models. Predictive Model Markup Language, or “PMML,” 17 00:01:35,350 --> 00:01:39,290 was the first such standard, based on XML. 18 00:01:39,290 --> 00:01:44,740 It was created in the 1990s by the Data Mining Group, a group of companies working together 19 00:01:44,740 --> 00:01:51,409 on the open standards for predictive model deployment. IBM and SPSS were among the founding 20 00:01:51,409 --> 00:01:54,299 members of the Data Mining Group. 21 00:01:54,299 --> 00:01:58,430 PMML 4.4 was recently released. 22 00:01:58,430 --> 00:02:04,229 It includes 17 statistical and machine learning models and many data transformations, built-in 23 00:02:04,229 --> 00:02:10,500 functions, ways to combine multiple models together, and other features. This standard 24 00:02:10,500 --> 00:02:17,140 is widely known and used. The products we looked at earlier -- Watson Studio, IBM SPSS 25 00:02:17,140 --> 00:02:24,760 Statistics, IBM SPSS Modeler -- enable users to export most models in PMML. 26 00:02:24,760 --> 00:02:32,209 In 2013, a demand for a new standard grew, one that did not describe models and their 27 00:02:32,209 --> 00:02:38,730 features, but rather the scoring procedure directly, and one that was based on JSON rather 28 00:02:38,730 --> 00:02:47,500 than XML. This led to the creation of Portable Format for Analytics, or PFA. PFA is now used 29 00:02:47,500 --> 00:02:51,840 by a number of companies and open source packages. 30 00:02:51,840 --> 00:03:00,489 After 2012, deep learning models became widely popular. Yet PMML and PFA did not react quickly 31 00:03:00,489 --> 00:03:06,439 enough to their proliferation. The need for a standard intermediate representation was 32 00:03:06,439 --> 00:03:12,709 amplified by the wide variety of emerging deep learning frameworks and specialized hardware. 33 00:03:12,709 --> 00:03:20,459 In 2017, Microsoft and Facebook created and open-sourced Open Neural Network Exchange 34 00:03:20,459 --> 00:03:26,849 , or “ONNX.” Originally created for neural networks, this format was later extended to 35 00:03:26,849 --> 00:03:30,459 support “traditional machine learning” as well. 36 00:03:30,459 --> 00:03:35,519 There are currently many companies working together to further develop and expand ONNX, 37 00:03:35,519 --> 00:03:42,019 and a wide range of products and open source packages are adding support for it. 38 00:03:42,019 --> 00:03:47,751 Watson Machine Learning is IBM’s commercial offering designed for model deployment. It 39 00:03:47,751 --> 00:03:53,349 supports deployment of models built with most open source packages, as well as those expressed 40 00:03:53,349 --> 00:04:02,359 in PMML or ONNX. It also supports deployment of IBM SPSS Modeler streams and Modeler flows 41 00:04:02,359 --> 00:04:08,959 from Watson Studio. Deployment can be done using a graphical interface or Python code, 42 00:04:08,959 --> 00:04:14,409 and can be for online scoring through a REST API or batch scoring. 43 00:04:14,409 --> 00:04:19,650 Watson Machine Learning helps integrate a deployed model into applications in the form 44 00:04:19,650 --> 00:04:24,510 of code snippets in several programming languages. 45 00:04:24,510 --> 00:04:30,110 In this video, you’ve learned how open standards and Watson Machine Learning can help users 46 00:04:30,110 --> 00:04:33,630 to deploy their models into various application. 47 00:04:33,630 --> 00:04:40,760 Next we’ll talk about AutoAI and OpenScale, two advanced Watson Studio features that help 48 00:04:40,760 --> 00:04:43,820 to further simplify a data scientist’s work.