Most companies think of machine learning as something super complicated, expensive, and requiring serious expertise. And if you intend to create a new Netflix recommendation system, it is. However, the trend of turning everything into a service has affected this complex area as well. It is possible to start an ML project from scratch without much investment, and it will be the right decision if your company is new to data science and wants to start by solving the simplest tasks.
One of the most inspiring stories about ML is the one about a Japanese farmer who decided to automatically sort cucumbers to help his parents in this tedious job. Unlike big corporations, this guy had no experience in machine learning and no big budget. However, he was able to master TensorFlow and apply deep learning to recognize different classes of cucumbers.
With cloud-based machine learning services, you can start building your first working models, making valuable insights from predictions even with a small team. We’ve already talked about machine learning strategy. Now let’s take a look at the best machine learning platforms on the market and talk about the infrastructure decisions you need to make.
In this article, we’ll first provide a brief overview of the major machine-learning-as-a-service platforms from Amazon, Google, Microsoft, and IBM, and then compare the machine learning APIs these companies support. It’s worth noting that the overview will not provide comprehensive instructions on how and when to use these platforms, but rather information on what to look for when reviewing their documentation.
Amazon machine learning
Amazon Machine Learning for predictive analytics is one of the most automated ML solutions on the market, best suited for operations where deadlines are very important. The service can load data from multiple sources, including Amazon RDS, Amazon Redshift, CSV files, and so on. All data preprocessing operations are performed automatically: the service determines which fields are categorical and which are numeric, and does not ask the user to choose how to further preprocess the data (dimensionality reduction and data whitening).
Amazon ML prediction capabilities are limited to three options: binary classification, multiclass classification, and regression. Meanwhile, the Amazon ML service does not support teacherless learning techniques, and the user must select a target variable to partition it in the training dataset. In addition, the user does not have to know the machine learning techniques, as Amazon selects them automatically after examining the provided data.
It’s worth remembering that after 2021, Amazon no longer updates either the documentation or the Machine Learning platform itself. The service still works, but does not accept new users. This is due to the fact that SageMaker and its related services are superior to AML in every way, and essentially provides users with the same functionality.
Predictive analytics can be used in the form of real-time or on-demand data using two separate APIs. The only thing to consider is that Amazon currently seems to be emphasizing its more powerful ML services, such as the SageMaker described below.
This high level of automation is both an advantage and a disadvantage when using Amazon ML. If you need a fully automated but limited solution, AML meets your expectations. Otherwise, choose SageMaker.
SageMaker
SageMaker is a machine learning environment designed to simplify the work of a data scientist. It provides tools to quickly create and deploy models. For example, it has a Jupyter notebook to simplify data exploration and analysis without fiddling with server management.
In 2021, Amazon launched SageMaker Studio, the first IDE for machine learning. This tool introduces a web-based interface that allows you to run all ML model training tests in a single environment. All development techniques and tools, including notebooks, debugging tools, data modeling, and automated generation are available in SageMaker Studio.
SageMaker’s built-in methodologies overlap heavily with the ML APIs offered by Amazon, but here datascientists can experiment with them and use their own data sets.
If you don’t want to use them, you can add your own techniques and run models with SageMaker, applying its deployment features. Or you can integrate SageMaker with TensorFlow, Keras, Gluon, Torch, MXNet, and other machine learning libraries.
In general, Amazon’s machine learning services provide enough freedom for both experienced data scientists and those who need to perform tasks without delving into data preparation and modeling. These services can be a solid option for companies that are already using Amazon cloud services and have no plans to move to other cloud providers.
The popularity of DevOps in the software development community has spawned the concept of “MLOps.” DevOps is a software development methodology that involves merging development and operations teams to optimize software development processes through a system of short and fast releases. It is realized by applying a high level of automation to routine tasks. MLOps, in turn, applies the same principles to machine learning, which has led to the emergence of automated data management, model training/deployment, and monitoring.
