September 2017 Summaries

The latest version of Redis, version 4.0, is now available, offering several significant enhancements including improved performance, memory usage, and bug fixes. The new release introduces features such as Redis Modules, Async non-blocking delete operations, and Mixed RDB-AOF format, which enables faster rewrites and reloads. Additionally, Redis Enterprise now supports Redis 4.0, allowing customers to explore the new version immediately, while Redise Pack 5.0 offers a preview of the new capabilities in Redis 4.0 engine, including geo-distributed deployments, real-time indexing, and native JSON processing. The latest release is designed to provide improved performance, memory usage, and bug fixes, making it an exciting update for developers and users alike.

The Redis team is planning for RedisConf 2018, which will take place in San Francisco at the end of April next year. In the meantime, the team has planned various events in September and October, including conferences, meetups, and community activities. These events include the Strata Data Conference, Strange Loop 2017, and meetups organized by Redis France and Seattle Redis Meetup groups. The team will also be represented at Cloud Foundry Summit in Switzerland, DockerCon EU in Denmark, Oracle Open World, and Spark Summit Europe in Ireland. Various speakers from the Redis engineering team will present on topics such as using Redis with machine learning architectures, choosing a data store for microservices, and using Redis to serve predictive models trained in Spark. The events are designed to provide opportunities for users to learn more about Redis and its capabilities.

The Redis ecosystem has introduced the RediSearch module, which provides three main features: full-text searching, general-purpose secondary indexing, and a suggestion engine. The module utilizes its own datatype and the built-in Redis data types, making it a solution that resides with Redis. With RediSearch, users can index text without pre-processing and find matches automatically, including numeric and geo indexes on individual items. The module also provides a suggestion engine for auto-complete-like services. After loading the module, users can create an index and add documents to it using the FT.CREATE and FT.ADD commands. These commands specify fields in a [fieldname] [value] repeating pattern, allowing users to store data with RediSearch's native datatype. The stored data is then accessible through various keys, including hashes, numeric indexes, geo sets, and textual indexes, each with their own specific type and characteristics.

This post discusses the benefits of using Redis for real-time predictive engines in machine learning applications. It highlights that most ML toolkits focus on learning, leaving the problem of building a predictive engine to developers. Redis fills this gap by providing a familiar and full-featured data store for building applications. The advantages include simple and powerful data management, reduced operational complexity, and fast throughput due to in-memory data storage. Additionally, Redis already integrates with most companies' tech stacks, making it easier to manage and scale compared to custom services.

E-mail marketing, despite its challenges, remains an effective way to reach customers. One technique that provides dynamic control after sending emails involves using Redis and a Node.js server alongside regular email tools. By pointing badge image URLs and item URLs at intermediate URLs, marketers can calculate the "hotness" and popularity of items in real-time, allowing for more effective communication with customers. This technique is efficient in terms of storage space and provides rich data on customer interactions. However, it requires a chatty system with multiple moving parts and may present challenges under heavy load. Redis Enterprise can provide high-availability and clustering to mitigate these issues. The same technique can be applied to other platforms for dynamically changing images based on data changes.

Fast data ingest and processing is crucial for big data applications in fields like IoT, e-commerce, security, communications, entertainment, finance, and retail. Redis offers a solution to these challenges with its high performance, variety of data structures, and efficient message broker capabilities. It can handle over one million read/write operations per second with sub-millisecond latencies on commodity cloud instances. Redis' publish/subscribe feature allows it to act as an efficient message broker between geographically distributed data ingest nodes. For larger datasets, Redis Enterprise combines RAM and Flash memory for cost-effective performance at scale.

The article explores the use of decision trees in machine learning for classification and regression problems. It discusses how decision trees model a sequence of rules as a binary tree, where interior nodes represent splits or rules and leaves represent classifications or values. The article then applies this concept to build a Titanic survival predictor using Python's scikit-learn package and Redis. The data is preprocessed by removing sparse columns and encoding categorical features, and a decision tree classifier is built with a maximum depth of 10. The article demonstrates how to load the decision tree into Redis using the ML.FOREST.ADD command and evaluate predictions using the ML.FOREST.RUN command. Finally, it compares the results of Redis' predictions with those of scikit-learn, showing that they are identical in all cases, including misclassifications.