We review PredicSis AI, PredictionIO and Seldon, three analytic services based on predictive APIs
PredicSis AI, PredictionIO and Seldon are three platforms using APIs to collect and manage large volumes of aggregated data that is used to create predictive models. Here we explain how they work and their main features.
The value of analytics as a key element in business decision-making is beyond question. Companies have gradually incorporated analytics and the concept of metrics of success into their decision-making processes, although it is something not yet universally extended. There are still many companies that base all their decisions on the intuition of their senior and experienced professionals or directly on the decisions of their CEOs. Within the analytics field there are some interesting tools that go a step further: they base their analyses on the power of predictive APIs, which generate value through machine learning and continuous learning.
Predictive APIs are enabling the accessible use of machine learning. They detect data patterns and assign the probability that a future fact belongs to that particular pattern, and they generate an efficient prediction model. These APIs enable developers to create models using historic information in financial institutions to detect fraud, in large corporations to control price policies, in electric companies to anticipate demand, etc. There are numerous examples.
There is analytical software in the market that uses machine learning and data to meet business goals:
PredicSis AI is a platform that is essentially aimed at certain technical, analysis and business profiles in any company: company executives, software engineers, account executives, business analysts, DevOps teams and data scientists. To test it, any company can request a trial by filling out a form on their website. There are four steps in the process that establish business predictions:
● Implementation of indicators: PredicSis AI can be used to gather a large amount of aggregated data from many different sources. This large set of information is the basis from which decisions will be made.
● Construction of segments: it is essential for professionals in general, but much more for those who have no technical training, to establish optimal segments to discover the real relationships between data, especially if these relationships come from business decisions. It shows a large volume of information using simple and relevant ideas.
● Prediction model: in this phase, PredicSis AI generates a formula or model that is used to predict future behavior. In this process, it is important to control the deviation of this prediction. If there is a large deviation of values, business decisions may be wrong.
● Final prediction: based on the three previous steps, the model uses the most recent data to generate the prediction using the target audience. The predictive model can improve the performance of campaigns and processes.
PredicSis AI is machine learning software that uses supervised learning algorithms to create models. Professionals who use the platform can access it through a graphical interface in a web browser, as if browsing the internet, or through a programming interface using their SDK in Python or through calls to their API REST.
Apache PredictionIO is an open source machine learning server built on the top of the stack so that developers and data scientists can create predictive engines and models with business goals. If you look at the image above, PredictionIO can include prediction models and machine learning processes in a mobile application. Doing this from scratch requires a lot of effort, time and higher costs in training an algorithm. PredictionIO is nothing more than a LAMP server for the analysis of data through predictive models, which is responsible for the entire cumbersome process of managing the algorithms, their training, their implementation at the top of an application where they are executed, the different dependencies, etc.
This service has the following characteristics:
● Responds to dynamic queries in real time.
● Unifies data from different platforms in batches or in real time.
● It has machine learning libraries and data processing such as Spark MLLib and OpenNLP. Spark MLLib contains logistic regression algorithms and support vector machines (SVM), Bayesian regression tree models; least squares techniques; Gaussian mix models; analysis of K-means clustering; latent dirichlet allocation (LDA); singular values decompression (SVD); principal component analysis (PCA); linear regression; isotonic regression, etc.
● Facilitates data infrastructure management.
Apache PredictionIO can be installed as a complete machine learning stack, with Apache Spark, MLLib, HBase, Spray and Elasticsearch.
Seldon is a predictive platform that provides content recommendations that are built on a Kubernetes cluster. Kubernetes is an open source system created by Google to program the deployment, scaling and monitoring of applications packaged in containers, hosted in the cloud and in need of computing. This system is present in search engine projects as relevant as Google Drive or Google Maps. The packaging provided by Kubernetes allows users to take the applications to any platform and execute them, which can be Amazon Web Services, Google Cloud Platform or Microsoft Azure.
Some of the most important Seldon characteristics are:
● Content and product recommendations: Seldon allows users to capture and record user actions through their REST API, and then use that information to deliver personalized recommendations to other users. Seldon's infrastructure consists of a set of layers: a real-time layer, responsible for managing real-time predictive API requests; the storage layer, which manages the storage of the various components of the infrastructure; and the statistics layer, which monitors and analyzes the system in operation.
● Making predictions: the aggregated data, which is the basis of any predictive model, is sent to the platform via the REST API in real-time. The application development interface collects data from multiple sources to create the predictive model. Usually, these data are sent in JSON format and then a data modeling process is created using algorithms, among other reasons because JSON is not the best format to create machine learning models.
Sign up to the BBVAOPEN4U newsletter and receive tips, tools and the most innovative events directly in your inbox.