One of the key technologies for innovation in companies is machine learning (ML): ML supports the optimal distribution of company resources, making data-based decisions in real-time or providing decision support. At the same time, the use of ML solutions requires extensive knowledge in data science. Experts in this field are in demand.
To not have to do without urgently needed technical innovations, the available data experts should be supported in their work in the best possible way. Self-service ML platforms fill the gap here. These platforms aim to enable employees with less prior knowledge to develop ML models, support data scientists in their work and simplify the use of ML models in general in companies.
A carefully selected self-service ML platform can significantly positively impact a company’s development in ML. However, many different variants on the market make it difficult to choose the right solution. What Matters When Choosing Self-Service ML Platforms? An overview of the key features of self-service ML platforms:
Consideration Of Goals And Requirements
Organizations considering adopting a self-service ML platform typically do one or more of the following:
- Increase the productivity of their teams working on ML use cases by improving use case lifecycle management, accelerating the trial and error process, and transforming the use case development process to be less error-prone.
- Reduction of the requirements regarding the qualification of employees for the development of ML uses cases to increase the proportion of data-based decisions and processes in the company.
- Facilitate collaboration between team members by making self-service ML platforms accessible to different skill levels within the team.
- Assist in going live and managing ML use cases to realize desired business benefits.
Various self-service ML tools are available with features covering some or all of these goals. It is beneficial to choose a self-service ML platform that meets all objectives. The development teams will also appreciate such a platform. However, the most powerful platforms are also the most expensive, especially ongoing license fees. Therefore, companies must ask themselves what their goals are and to what extent the self-service ML platform should contribute to focus on the most important requirements during the evaluation phase.
The Most Important Features
Self-service ML platforms can support key needs in ML use case development.
Model Development:
In the first step of the development cycle, ML models must be trained using datasets relevant to the use case. Self-service ML platforms provide ready-to-use and parameterizable models that can be used through a graphical user interface. Most media support learning at least one out-of-the-box model at a given time.
More mature self-service ML platforms offer a recommendation engine that shows the customer which ML models suit a specific use case. It is also possible to train several ML models in parallel for one use case.
High-end self-service ML platforms offer a feature called “Auto ML”. The appropriate algorithm for learning the ML model for the application is automatically selected here. In addition, they can contain functions for individually developed ML models in various programming languages such as R, Python or Julia.
Model Evaluation:
To evaluate the performance of ML models, low-end self-service ML platforms provide reports with corresponding goodness measures. Self-service ML platforms enable a direct comparison between all models taught in parallel so that the best model can be selected for the complementary use case. High-end platforms with Auto-ML capability will automatically choose the optimal model.
Visualization:
As part of the development cycle, it is important to analyze the raw data and the prepared input data and get a clear picture of the learning outcomes of the ML model. The visualization capabilities of self-service ML platforms are therefore crucial. The platforms should be able to present the final results to allow quick decision-making.
The functionalities of the various platforms range from a simple graphical representation to implementations similar to self-service BI tools, which also enable complex issues to be visualized quickly.
Data Cleaning And Preparation:
The training data for an ML model must be obtained from different sources and converted into suitable data sets. Self-service platforms provide similar capabilities as ETL tools and additional ML-specific data cleansing and preparation capabilities.
Platforms with auto-ML capabilities can also partially or fully automate these tasks. In addition, self-service ML platforms offer the option to use code in data cleaning and preparation and offload calculations to a database or cluster for faster processing.
Collaboration:
Choosing a self-service ML platform requires very careful consideration of the skills available in the team. To improve cooperation, some self-service ML platforms have built-in collaboration capabilities. Commonly offered features are version control, sharing of work products, commenting features, to-do lists, using no-code/low-code or code, and knowledge sharing features.
While data scientists and data engineers can collaborate through code and code repositories, business analysts are more likely to work with no-code, low-code, and visualization capabilities. When data scientists, data engineers, and business analysts work together, the collaboration features should be appropriate for all roles involved.
Operationalization Of ML Use Cases:
The operationalization of ML use cases is a complex task. ML use cases must be integrated into business processes and the results made available for analysis and reporting. Self-service ML platforms provide capabilities to expose ML models as an API. In this way, they can be used in real-time by third-party systems. The ML models are exported to be used within the existing IT infrastructure. Another example is the scheduled automatic execution of an ML model to provide results to a subsequent system automatically.
Self-service ML platforms with these capabilities enable ML models to be integrated into business processes and reduce the need for administration. To achieve a business advantage, the ML models must be integrated into the technical processes, which results in additional expenses in the supplying and consuming systems.
Monitoring:
This point must be considered from two perspectives: Firstly, it is about technical tracking, i.e., observing the status of the various technical components or jobs. With every operationalization approach (as mentioned above), the corresponding monitoring functionalities must also go hand in hand.
Second, monitoring for business metrics is also important, and some more mature self-service ML platforms also support it. These work with a closed-loop approach to finding out whether applying an ML model has proven helpful or not. However, one must remember that such KPI monitoring must be embedded in the technical processes and cannot be solved alone by the self-service ML platform.
Enterprise Maturity:
The requirements for self-service ML platforms in large companies differ from those of small companies in terms of security, access management, governance and availability. The more employees and teams from different business areas or projects are involved, the higher the effort for administration and protecting the company’s valuable assets. These requirements are met by enterprise-grade self-service ML platforms that offer, for example, Lightweight Directory Access Protocol (LDAP) integration, multi-user security, user management, auditing capabilities, fault tolerance and scalability.
Conclusion
The range of self-service ML capabilities offered and their realizations are vast, and the full list of capabilities is even longer. This assesses self-service ML platforms as a complex task, requiring evaluation of requirements that also cover:
- Strategy: The portfolio of ML use cases to be realized, the associated business cases, and the planned time to market are important parameters.
- Organization: The organizational structure, the range of qualifications within the ML development team(s) and, of course, the business model plays an important role here.
- Architecture: Architectural guidelines should be considered for the integration, and smooth operation of such a platform within an organization’s IT environment, as well as the coverage of quality attributes such as scalability, extensibility, security, portability, fault tolerance and availability.
These aspects are particularly important to guarantee that the selected self-service ML platform meets the business’s specific needs. Otherwise, ML use cases that have already been developed would be “trapped” on the self-service ML platform and would have to be redeveloped with additional effort. In addition, there are already costs due to ongoing license fees. This risk can be countered with the support of external expertise to identify the right self-service machine learning platform for your own company and to ensure successful implementation. Furthermore, in developing the initial machine learning use cases, so-called “on-the-job training” can be carried out with external support, and initial success can be achieved more quickly.
ALSO READ: AI From The Cloud – A Look At The Practice
