This article details Qunar's migration from Prometheus to VictoriaMetrics for large‑scale container monitoring, covering the shortcomings of Prometheus at massive data volumes, the architectural choices made, performance improvements achieved, and future optimization plans.
The article details a comprehensive presentation on Alibaba Health's pharmaceutical e‑commerce sales forecasting, covering supply‑chain challenges, the evolution of time‑series prediction methods, a full data‑to‑model pipeline, change‑point detection, handling imbalanced data, multi‑model fusion, and specialized seasonal and long‑sequence forecasting techniques.
This article explains the internal storage layout, on‑disk and in‑memory data structures, and the query execution flow of Prometheus V2, illustrating how blocks, chunks, WAL, indexes and postings are organized and accessed to serve time‑series queries efficiently.
This article provides an in‑depth analysis of Prometheus V2's storage architecture, detailing the on‑disk block layout, chunk and index formats, the inverted index structure, memory representations, and the step‑by‑step query process that locates matching time‑series data.
This article introduces the tsfresh Python package, explains why traditional machine‑learning models struggle with time‑series data, and demonstrates how tsfresh can automatically generate and select hundreds of useful features—including statistical, nonlinear, and signal‑processing metrics—while supporting big‑data frameworks such as Dask and Spark.
This tutorial explains how to convert single‑variable and multivariate time‑series data into supervised learning formats using pandas shift() and a custom series_to_supervised() function, covering one‑step, multi‑step, and sequence forecasting examples with complete Python code.
This article presents a threshold‑free anomaly detection framework for streaming time series that combines an LSTM‑based baseline module with an unsupervised detection module, detailing the architecture, training process, data preprocessing, and experimental results that demonstrate superior accuracy and F1 scores.
This article explains how to identify, order‑select, estimate parameters, validate, and forecast ARMA time‑series models, covering the Akaike Information Criterion, various estimation techniques, and diagnostic tests such as the Ljung‑Box test, with practical Python implementation guidance.
Grey system theory provides a framework for analyzing, modeling, and predicting systems with incomplete information, using minimal data and accumulation techniques to improve forecast accuracy, as demonstrated through a sales data case study with Python implementation.
This curated list groups recent articles on change and prediction models, covering topics such as war dynamics, population, infectious diseases, differential equations, regression, time‑series analysis, and various machine‑learning techniques, providing quick access to each resource.
This article presents a comprehensive overview of using Elasticsearch as a time series engine, covering its motivations, challenges, key features, Alibaba Cloud TimeStream optimizations such as columnar storage, LSM structures, downsampling, and integration with Prometheus and Grafana, while also discussing performance and cost considerations.
This guide explains the limitations of ARIMA‑type models for handling changing variance, introduces ARCH and GARCH as solutions, and walks through Python implementations—including data generation, model fitting, and forecasting—complete with code snippets and visualizations.
A time series is a chronologically ordered set of interrelated data points whose analysis involves studying its development patterns and forecasting future behavior, with classifications based on dimensionality, continuity, statistical properties such as stationarity, and distribution types like Gaussian or non‑Gaussian.
This article explains why Elasticsearch is being adapted as a time‑series engine, outlines its unique characteristics and challenges such as high query complexity and storage cost, and introduces Alibaba Cloud’s TimeStream solution with optimizations like index settings, compression, down‑sampling, and Prometheus integration.
This article provides a comprehensive overview of Vivo's server‑side monitoring system, detailing its architecture evolution, data collection pipelines, OpenTSDB storage design, alerting mechanisms, and comparisons with other mainstream monitoring solutions, offering practical guidance for technology selection and implementation.
The article describes the motivation, requirements, and architectural design of a highly available, scalable, low‑cost distributed time‑series database built on Mimir, detailing write and read paths, multi‑tenant isolation, compaction, and the performance and cost improvements achieved after deployment.
The GM(1,N) prediction model extends the classic GM(1,1) approach to multiple indicator variables by applying accumulated generating operations, forming a first‑order differential equation, converting it to a discrete model, and using least‑squares estimation to derive prediction values for each variable.
This guide explains the theory behind Vector Autoregression (VAR) models, outlines the complete modeling workflow—including data preparation, stationarity and cointegration testing, lag order selection, parameter estimation, stability diagnostics, and impulse‑response and variance‑decomposition analysis—and provides a full Python implementation with code examples.
This article explains why residuals of an ARMA model should be white noise, how to use the Q‑test to detect autocorrelation, and how AIC and BIC criteria balance model fit against complexity for selecting the most appropriate ARMA specification.
This article explains how to recognize whether a stationary random time series follows a pure AR, pure MA, or mixed ARMA process by examining the patterns of the autocorrelation function (ACF) and the partial autocorrelation function (PACF).
This article explains the concept of differencing for extracting deterministic information, outlines when to use first, second, or seasonal differencing, introduces the ARIMA(p,d,q) model and its special cases, and provides a complete Python example with code and visualizations.
This article explains the recursive forecasting formulas for ARMA and MA(q) time‑series models, showing how forecasts depend only on past observations, how model invertibility ensures stability, and how estimated parameters are used in practical prediction.
This article explains the process of constructing ARMA models, covering model identification, order selection using the AIC criterion, parameter estimation (including Python implementation), and diagnostic testing such as Ljung‑Box, before demonstrating how to generate forecasts from the fitted model.
This article explains the three main types of stationary time‑series models—AR, MA, and ARMA—detailing their definitions, back‑shift operator notation, polynomial representations, and the essential stationarity and invertibility conditions required for valid modeling.
This article explains the concept of wide‑sense stationary time series, introduces essential definitions such as mean, variance, autocovariance, autocorrelation, and white‑noise processes, and derives the Yule‑Walker equations that characterize linear stationary models.
Learn a step-by-step approach to predict seasonal time series—such as product sales or climate data—using the seasonal coefficient method, illustrated with a quarterly refrigerator sales case study and a complete Python implementation that computes next year's quarterly forecasts.
This curated list groups recent articles on change and predictive models, covering topics such as war dynamics, population, epidemic spread, differential equations, regression, time‑series analysis, machine learning classifiers, and grey‑prediction techniques, providing students with ready references for diverse modeling approaches.
This article explains why downsampling is essential for handling massive time‑series data in Prometheus, describes the aggregation rules and intervals, compares ARMS Prometheus' implementation with other solutions, and shows performance and accuracy results that demonstrate significant query speed improvements.
This article explains the functional, nested nature of PromQL, its expression types, how queries are parsed and evaluated over time, and the differences between instant and range queries, providing code examples and visual insights for better monitoring with Prometheus.
This article walks through why accurate demand forecasting is critical for retailers, shows how to prepare and visualize sales data, demonstrates building a store‑item model with Databricks DDI and Facebook Prophet, and explains scaling the model to predict every product across all stores, highlighting performance metrics and practical tips.
This guide explains how Hyperscale (Citus) extends PostgreSQL with distributed tables, enabling developers to start on a single node and seamlessly scale to multiple nodes for high‑throughput SaaS, real‑time analytics, and time‑series workloads across many industry verticals.
The article describes JD Search's machine‑learning alert system that combines offline and real‑time training, FFT‑based periodic detection, Prophet forecasting, and DBSCAN anomaly clustering, and explains architectural design, data preprocessing, model optimization, and distributed deployment to improve alert accuracy and response speed.
The article describes JD Search’s machine‑learning‑based time‑series forecasting and anomaly‑detection platform, detailing its overall architecture, offline and real‑time training pipelines, FFT‑based periodicity detection, Prophet forecasting, DBSCAN outlier detection, and distributed optimizations such as Alink integration and load‑balancing strategies.
This article explains how Prometheus’s time‑series database handles massive monitoring data, from basic concepts and query examples to storage engine design, indexing strategies, and powerful data computation techniques such as recording rules.
This tutorial explains how to convert single‑variable and multi‑variable time‑series data into a supervised‑learning format using Pandas' shift() function and a custom series_to_supervised() helper, covering one‑step, multi‑step, and multivariate forecasting examples with complete Python code.
This guide shows how to combine PostgreSQL declarative time‑range partitioning with Citus distributed sharding to create a scalable time‑series database, covering table creation, distribution, automatic monthly partition generation, columnar archiving, maintenance with pg_cron, and performance benefits such as faster inserts, queries, and data expiration.
This article details Alibaba's exploration of intelligent supply‑chain forecasting, covering scenario classification, three generations of prediction algorithms, the self‑developed Falcon model, performance evaluation, and real‑world cases such as Double 11 and live‑streaming, highlighting challenges and practical solutions.
This article explains essential time series preprocessing techniques—including data sorting, handling missing values with interpolation methods, applying rolling averages, Fourier transform denoising, and detecting anomalies using rolling statistics, isolation forests, and K‑means clustering—illustrated with Python code on the AirPassengers and Google stock datasets.
This tutorial explains how to work with time‑series data in Python, covering dataset preparation, resampling techniques, Prophet forecasting, LSTM‑based prediction, and hierarchical clustering, while illustrating each step with visual examples and code snippets.
This article details the background, research, architecture, performance testing, and deployment of a comprehensive monitoring system that leverages Prometheus, Grafana, and M3DB to provide flexible metric collection, automatic dashboard generation, and a custom alerting service for large‑scale business services.
This article explains a NeurIPS 2021 paper that introduces a learnable autocorrelation correction for neural network time‑series models, shows the underlying theory, provides concise PyTorch code implementing the adjustment, reports a ~17% average performance gain across datasets, and lists additional practical tricks for time‑series forecasting.
This article explains how mixing a Poisson process with a discrete k‑state hidden Markov model creates a Poisson HMM that captures autocorrelation in integer‑valued time‑series, detailing the model formulation, prediction via expectation over states, and parameter estimation using MLE or EM.
This article examines the state of industrial AI, discussing data and model challenges, the multidisciplinary talent needed, the DIKW framework, typical AI workflows, edge‑cloud architecture, real‑time processing tools, time‑series storage, service design patterns, and practical recommendations for deploying AI in manufacturing.
The article explains why the Internet of Things generates massive, diverse data streams that require specialized databases, outlines key selection criteria, describes common IoT data types, and reviews several open‑source databases—InfluxDB, CrateDB, MongoDB, RethinkDB, SQLite, and Cassandra—highlighting their strengths for IoT workloads.
This tutorial walks through the full Python-based ARIMA modeling process for time‑series analysis, covering data loading, stationarity and white‑noise tests, model order selection, parameter estimation, diagnostic checks, and future forecasting with detailed code examples.
This article describes how a financial company leveraged deep‑learning sequence models to automatically extract features from massive time‑series data, improving risk‑assessment models and operational efficiency through a unified framework that includes data preprocessing, embedding, field and item aggregation, and end‑to‑end deployment.
Snuba is a ClickHouse‑based service that offers rich data models, fast Kafka ingestion, and a rule‑based query optimizer for Sentry, replacing the original Postgres‑Redis stack and supporting multiple time‑series datasets, streaming and point‑in‑time queries with configurable consistency guarantees.
This article explains a simple investment timing strategy that adjusts regular contributions based on daily northbound capital flows, walks through the financial concepts, provides complete Python code using Tushare to fetch and process the data, and shows back‑tested results and practical tips.
Boussole is Shopee’s real‑time analytics engine that transforms each dimension into key‑value pairs stored primarily in HBase, pre‑aggregates selected dimension combos, hashes metrics and tags, executes distributed PromQL queries with a CockroachDB‑inspired executor, applies Delta‑of‑Delta compression and point‑capping, and continues to evolve with adaptive pre‑aggregation and new storage models to maintain millisecond latency for massive multi‑dimensional analysis.
The article explains how Prometheus, a data‑driven monitoring system, handles massive time‑series data using its TSDB storage engine, detailing concepts, query examples, storage characteristics, indexing mechanisms, and the benefits of pre‑computing rules for efficient monitoring at scale.
This article explains how hobbyists can leverage time‑series databases, Raspberry Pi, MQTT and Slack alerts to continuously monitor and control temperatures during beer fermentation and barbecue smoking, providing practical setup steps, alert thresholds, and visualization tips for reliable results.
The September 2021 DB‑Engines ranking shows major declines for Oracle, MySQL and SQL Server, rapid gains for Snowflake and MongoDB, and highlights emerging time‑series and key‑value databases like InfluxDB, Redis, and ClickHouse reshaping the database market.
This article reviews the evolution of end-to-end deep learning models, describes how they were adapted and optimized for a real‑estate broker churn‑warning scenario, and presents experimental results showing significant improvements in AUC, KS and lift over traditional classifiers.
The paper presents a graph convolutional network approach that leverages multi‑task learning and spectral graph convolutions to forecast shared‑bike inflow, outflow, and demand gaps across a city’s non‑Euclidean parking network, demonstrating improved accuracy over traditional time‑series baselines while noting scalability and directional graph limitations.
This article introduces a curated selection of lesser‑known Python libraries for data‑science tasks—including data acquisition, date‑time handling, imbalanced‑learning, fast keyword extraction, fuzzy string matching, time‑series modeling, 3‑D visualization, web‑app building, and reinforcement‑learning—providing installation commands and concise usage examples.
This article examines the explosive growth of IoT data, outlines the unique challenges of storing and querying massive time‑series datasets, reviews the evolution of time‑series databases, and explains how Alibaba Cloud's Lindorm TSDB leverages cloud‑native, multi‑model architecture to deliver high‑throughput, low‑cost, and scalable solutions for IoT, industrial and monitoring workloads.
This article explains how Prometheus turns complex monitoring into manageable tasks by introducing its core concepts, daily query examples, the advantages of its TSDB storage engine, and powerful data computation techniques that enable efficient large‑scale time‑series analysis.
This article demonstrates how to use Python's data‑analysis ecosystem (NumPy, pandas, statsmodels, etc.) together with Flask, MySQL, and ECharts to create a web‑based sales forecasting tool that automates data extraction, modeling, prediction, and visualisation.
Time series databases let you retain full historical records, enabling analysis, visualization, machine learning and automation across domains like finance, weather and IoT, and the article explains why they’re essential, how they differ from traditional databases, and how to start using them.
This article explains how JD’s JUST platform leverages the open‑source columnar database ClickHouse to store, query and analyze massive time‑series datasets, covering data modeling, lifecycle management, cluster architecture, write and query processes, scaling strategies and future enhancements.
This article outlines Ctrip's approach to time series forecasting, covering background, common tools such as factor‑based models, traditional statistical methods like ARIMA, and machine‑learning techniques including tree and neural networks, and shares practical experiences on data splitting, feature engineering, model stability, and evaluation.
This article presents a comprehensive overview of financial big‑data risk control models at Du Xiaoman, covering traditional scoring cards, AI‑driven time‑series and text processing, graph‑based networks, model interpretability, probability calibration, stability analysis, and the specific challenges introduced by the COVID‑19 pandemic.
This article explains the cause of InfluxDB’s max‑value‑per‑tag error when Prometheus remote‑writes high‑cardinality tags, analyzes why the built‑in memory index triggers OOM protection, and presents three practical solutions—including moving indexes to disk, storing high‑cardinality tags as fields, and filtering them before write—to ensure stable monitoring data persistence.
This article examines the challenges of performing time‑series similarity queries on massive datasets and presents three scalable solutions—partition‑based indexing, dimensionality‑reduction using MinHash, and a combined approach with Locality Sensitive Hashing—to reduce computation while preserving similarity accuracy.
The article explains why the massive, diverse data generated by the Internet of Things requires specialized databases, outlines key criteria for selecting a suitable database, and reviews several popular options such as InfluxDB, CrateDB, MongoDB, RethinkDB, SQLite, and Cassandra, highlighting their strengths for IoT workloads.
This article introduces time series fundamentals, classification, and challenges for internet businesses, then provides a step‑by‑step Python tutorial on ARIMA modeling—including data loading, stationarity testing, differencing, ACF/PACF analysis, AIC‑based order selection, model training, prediction, error evaluation, exogenous variable integration, and diagnostic checks.
This article introduces a curated collection of lesser‑known Python libraries for data‑science tasks—including wget, pendulum, imbalanced‑learn, flashtext, fuzzywuzzy, pyflux, ipyvolume, dash, and gym—detailing their purpose, installation commands, and concise code examples to help practitioners expand their toolkit.
This article introduces Prometheus' query language PromQL, explains instant and range vector selectors, label matching, offset handling, storage design, common functions and aggregation operators, and provides practical advice for efficient querying and avoiding performance issues.
This article provides a comprehensive tutorial on Prophet's key parameters, their meanings, and practical tips for tuning them—including growth, changepoints, seasonalities, holidays, and Bayesian settings—along with Python code examples for grid search and cross‑validation to improve forecasting accuracy.
This article introduces TimescaleDB—a PostgreSQL extension for time‑series data—explaining its core concepts, data‑model choices, architecture, native compression, and provides step‑by‑step installation instructions on a CentOS 7 environment.
This article details how Amap leverages deep‑learning techniques—including temporal and auxiliary feature engineering, multi‑stage RNN models, Wide&Deep architectures, and an Attention‑TCN approach—to accurately identify and handle expired points of interest, improving map freshness and user experience.
This article presents an intelligent inventory management solution that predicts product consumption using two time‑series algorithms—Facebook's Prophet and LSTM deep learning—detailing data sources, preprocessing, model configuration, evaluation metrics, and a comparative analysis of their performance and suitability.
The article explains how advertising inventory forecasting—predicting how many users will see a specific ad—poses challenges due to fluctuating traffic and user segmentation, and describes a new deep spatial‑temporal tensor factorization model that dramatically improves prediction accuracy, scalability, and robustness for large‑scale ad platforms.
This tutorial explains how to load a CSV file of financial time‑series data into pandas, perform basic exploratory analysis, compute absolute and relative changes, generate log returns, visualize results, and apply weekly or monthly resampling while avoiding look‑ahead bias.
The article presents a complete workflow for predicting next‑day API request volumes by exploring per‑minute traffic data, handling missing values, applying seasonal decomposition, training an ARIMA model on the trend component, and generating confidence intervals to flag anomalous spikes.
This article introduces Didi's open‑source monitoring system Nightingale, detailing its design, architecture, key improvements over Open‑Falcon, and how its flexible alerting and data handling capabilities support the full lifecycle of stability engineering in large‑scale operations.
The project develops a deep‑learning pipeline for Amap’s expired POI detection that integrates two‑year temporal trend features, industry and verification attributes, a variable‑length LSTM, a Wide‑Deep architecture, and a Wide‑Attention Temporal Convolutional Network, achieving higher accuracy and efficiency while outlining future macro‑and micro‑level enhancements.
Tencent leverages Elasticsearch at massive scale across log analytics, search services, and time‑series monitoring, addressing challenges of high availability, low cost, and high performance through kernel optimizations, resource‑aware throttling, cold‑data merging, rollup, caching, and open‑source contributions.
This article details how Tencent leverages Elasticsearch for log analysis, search services, and time‑series data, outlines the specific challenges faced in high‑availability and cost‑efficiency, and presents the comprehensive optimization techniques and future open‑source contributions that improve performance, scalability, and reliability.
This article explains the evolution of Alibaba DAMO Academy’s time‑series anomaly detection technology, detailing its application to both machine and commercial data, the challenges of diverse data types, the new robust statistical models, automatic data classification, parameter recommendation, and real‑world case studies demonstrating improved accuracy and stability.
A curated collection of Reddit‑highlighted 2019 AI research papers, covering theoretical advances, computer‑vision breakthroughs, unsupervised learning methods, and time‑series forecasting, with summaries, key contributions, and direct links to each paper.
The article reviews traditional, neural network, and open‑source time‑series forecasting methods, explains their strengths and limitations, and demonstrates how NetEase applies short‑term and long‑term prediction models such as Holt‑Winters, ARIMA, STL, Prophet, and LSTM to improve game monitoring and proactive alerting.
The iQIYI advertising team introduced an LSTM‑based deep‑learning model that forecasts inventory by normalizing data, clustering dimensions, and embedding fine‑grained holiday features, achieving significantly lower bias than their Adaptive‑ARIMA baseline and improving generalization while reducing training resources.
This article explains InfluxDB's Continuous Queries feature, covering configuration settings, basic and advanced syntax, execution timing, data range handling, result storage, and the use of RESAMPLE clauses with practical examples and code snippets.
This guide explains InfluxDB’s schemaless data types, best practices for designing tag‑and‑field schemas, and how to choose and configure its in‑memory or TSI indexes, including key parameters such as max‑series‑per‑database and max‑values‑per‑tag for optimal performance.
This article explains InfluxDB's retention policy components—duration, replication, and shard duration—clarifies the concepts of shards and shard groups, describes default configurations, offers recommendations for shard group duration, and outlines practical considerations for performance and data management.
The author describes encountering a Python 2.7 UnicodeEncodeError when Plotly attempts to open a browser for each generated time‑series chart, explains why the error occurs with non‑ASCII filenames, and shows how adding exception handling resolves the issue while processing weather data for thousands of cities.
This guide shows how to extract average API response times from a MySQL database using Java, process the data with a Python script, and generate an interactive time-series chart with Plotly, providing a practical method for long-term performance monitoring.
This article explains Elasticsearch X‑Pack's machine‑learning capabilities, covering supervised and unsupervised learning concepts, data preparation, task creation types, architecture components, data flow, result indices, and provides code examples for configuring and running ML jobs.
This article walks through the complete deployment of a time‑series monitoring solution using the TICK stack—installing and configuring InfluxDB, Telegraf, and Grafana—to collect, store, and visualize key metrics such as CPU, memory, network, and disk I/O for a big‑data platform.
To enhance estimated arrival times in navigation, this article analyzes the shortcomings of traditional historical average methods and proposes a machine‑learning solution using Temporal Convolutional Networks combined with dynamic and static feature engineering, demonstrating reduced bad‑case rates and better handling of seasonal patterns.
This article examines the limitations of Prometheus's original V2 time‑series storage, proposes a block‑oriented V3 architecture that tackles series churn, write amplification, and indexing inefficiencies, and validates the new design with extensive benchmarks showing dramatic reductions in memory, CPU, and disk usage.
This article provides a comprehensive overview of Prometheus, an open‑source monitoring and alerting system, covering its origins, key features, architecture, core components, metric types, data collection methods, service discovery, storage options, and query capabilities.
This comprehensive guide explores a wide range of anomaly detection techniques—including time‑series methods, statistical models, distance‑based approaches, tree‑based isolation forests, graph algorithms, behavior‑sequence Markov models, and supervised machine‑learning models—detailing their principles, formulas, and practical scenarios for detecting outliers in advertising, fraud, and system monitoring.
This article introduces the main business scenarios, system architecture, and core technologies of a time‑series data platform, covering data collection, processing, storage, analysis, and the specific features required for high‑performance, scalable, and reliable time‑series data management.
This article explains how to design and tune an Elasticsearch cluster for massive time‑series workloads, covering node role separation, storage strategies, routing, index splitting, hot‑warm architecture, cross‑cluster search and other performance‑boosting techniques.
This article introduces the concept of anomaly detection, outlines common application scenarios such as ELT pipelines, feature engineering, A/B testing, and fraud detection, and reviews various detection methods—including statistical models, machine learning, rule‑based logic, and density‑based techniques—while discussing practical implementation considerations.
This comprehensive guide explores anomaly (outlier) detection across diverse methods—including time‑series analysis, statistical tests, distance metrics, matrix factorization, graph approaches, behavior‑sequence modeling, and supervised machine‑learning models—highlighting their principles, formulas, and practical use cases such as fraud prevention and system monitoring.
This article outlines Alibaba Retail's demand forecasting workflow, describing the evolution from simple mean and time‑series models to machine‑learning and deep‑learning approaches, the incorporation of feature engineering, operational plans, and methods for estimating prediction uncertainty to support intelligent replenishment.
This article presents a detailed case study of building a purchase‑user prediction model by integrating Convolutional Neural Networks for feature extraction with Long Short‑Term Memory networks for time‑series forecasting, covering background, model structure, data augmentation, experimental results, and business impact.
