This is a rough field guide across the SAP wonderland of AI related software. Intended for developers, engineers, data science enthusiasts and practitioners who want to understand their ML model building options with SAP’s AI related software.
Opinions are my own.
Questions are welcome in the comments.
If you found this useful and know folks who will too, give us a like and share on social media. Thanks!
If you were to set yourself the task of identifying all the products we provide at SAP, I expect you would feel very much like Alice as she slipped into the rabbit hole down to Wonderland. It would be a strange land indeed with bots that talk, things that click and whir in the “cloud”, machines that explain what other machines do, and yet you would only be peeping at a little corner of Alice’s lovely garden.
Now, we would not want to confuse you as the Cheshire Cat confused Alice, with his rhetoric.
“Would you tell me, please, which way I ought to go from here?’
‘That depends a good deal on where you want to get to,’ said the Cat.
‘I don’t much care where –’ said Alice.
‘Then it doesn’t matter which way you go,’ said the Cat.
‘- so long as I get SOMEWHERE,’ Alice added.
‘Oh, you’re sure to do that,’ said the Cat, ‘if you only walk long enough.”
And so this blog, is an attempt to elucidate your options. If you are a developer, engineer or data scientist that is, particularly if you are looking to build and deploy a few ML models using our software. What you should use, when and why. Now if the Mad Hatter were throwing a tea party for you, the data adventurer, what would he serve? You’d see these laid out on impeccably white lace, with the SAP bright blue logo twinkling at you on the side.
- SAP Analytics Cloud
- SAP AI Business Services
- SAP HANA
- SAP Data Intelligence
“Speak English!’ said the Eaglet. ‘I don’t know the meaning of half those long words, and I don’t believe you do either!”
Well, no, we actually do. So let’s go through them one by one.
1. SAP ANALYTICS CLOUD
If you’re a business analyst, who understands your data and your problem statement very well, but don’t quite get that waltz with numbers that statisticians do, then this is what you’d pick.
Built keeping a “citizen data scientist audience” in mind, the Smart Predict feature of SAP Analytics Cloud supports you building predictive scenarios with little to no code. Simply load your data into the cloud, create any new features that might assist your prediction (this is where you can use functions akin to those in excel that perhaps count as code) and build your model. Currently, 3 of the most ubiquitous data science models – Classification, Regression & Time series forecasting – are supported in the context of tabular data (structured & semi-structured). In each model, you can define your target variable (the ground truth that you want predicted) and provide the explanatory variables (all those pieces of data with tell tales signs that relate to the truth). When the models run, Smart Predict runs a horse race of competing models and selects the one that offers the best performance for the final prediction. You get a list of stats related to the best model and a detailed summary of model predictions & behaviour in different segments of your data. Note here, that the actual ML model used is under the hood and its parameters cannot be customised. The classification & regression models use a variant of Ridge regression while the time series forecasting uses a combination of auto regression along with methods to identify trends & cycles in data.
Use when you aren’t particular about controlling the model being used, but care more about predictions that are good and can be used to augment your existing analytics dashboards.
2. SAP AI BUSINESS SERVICES & APPLICATIONS
If you have a business scenario that fits into one of the pre-packaged AI solutions we have, you’d wander over here because it means you don’t need to reinvent the wheel. Give the service a few inputs on what you need, and lean on the wisdom of data that’s piled over from 50 yrs of being in the business.
Here we offer, key services with pre-trained machine learning models that serve common business scenarios. See figure 1, for a brief on each of these.
Use when your use case ties in closely and can be defined within the ambit of the scenarios above, so you can drastically shorten the time to value.
3. SAP HANA
When it was released, SAP HANA was a revolutionary, first of its kind, in memory database. Many have come since us as far as “in memory” goes, but HANA is still one of its kind when it comes to the advanced analytics features it offers. Aside from building ML models, you can perform spatial data processing, streaming analytics and graph data processing. Since we’re focussing on building ML models in this blog post, I will only focus on PAL & APL (we love the alphabet soup, don’t we). The data scientists out there may care to note that HANA’s hardware has several TBs of working memory and a large number of CPUs, so those heavy ML models out there can power through on HANA’s steam.
a. PAL, or the Predictive Analysis Library, provides functions (both proprietary and open source) for building your models, called and executed as SQL procedures (through HANA SQLScript) or the Python module (called Hana-ml) or with a user friendly web IDE (called SAP HANA Application Function Modeler). The IDE has individual functions represented as graphical objects and the result is a flowchart that shows the movement of data through steps and algorithms. Functions can be dragged, dropped and parameterised in the GUI, making the code modular and easy to read. PAL’s algorithms are optimised for HANA and may run faster.
b. APL, or the Automated Predictive Library, uses pre-trained models for the typical classification, regression & time series problems. This is essentially the same engine that powers SAP Analytics Cloud’s Smart Predict, but is available in HANA in the form of a function.
Use when you want to custom build, deploy and maintain your ML models. Particularly when you want a single solution to both store your data (in-memory, no less), process it and run your predictive models on it. Or of course, if you have your business data sitting readily in SAP HANA and want avoid its unnecessary journey out for ML purposes. You’d work here with text / tabular data and you’ll need the data scientists for this one.
4. SAP DATA INTELLIGENCE
The metadata explorer offers an intuitive interface to play around with the data and understand the variable distributions. The integrated Jupyter lab environment makes the surroundings feel a little bit like home for the data scientists. The ML Scenario manager is where models can be trained, tested and deployed with pre-defined or custom operators. References to data sources enable audit trails for sensitive data. The data pipelining with a flowchart like GUI, allows for an agile, modular approach to coding. The ML Operations Cockpit helps deploy the models as a production service and embed the models into a business application, so they can be called when required.
Use when you want an enterprise ready data orchestration platform, to automate, maintain, scale and audit ML models. Data Intelligence will not discriminate your data on form and size, so bring in those audio files and images into the data lake, alongside the humble structured datasets. Yep, you need the data scientists here too.
This is not to say that this is a single choice game. Choose more than one solution if it suits your business needs. We throw you into the rabbit hole not to confuse you like the Cheshire Cat, but to give you options that we can help tailor to your varied needs.
So, hey data adventurer, what would you choose? I hope you enjoyed your tryst down here in this rabbit hole. And I hope I leave you, in Alice’s words, “Curiouser and curiouser!”
- For technological omniscience & help with vetting the messaging – Dr. Nicholas Nicoloudis, Daniel Dahlmeier, Suresh Kumar Raju, Nadine Hoffman, Badhri Sriram
- For creative effervescence & help with all the Alice in Wonderland quotes – Marvin Rougier
Other related AI technologies, where you can’t build ML models, but are cool anyway and fall under the “Intelligent technologies” umbrella.