Anomaly
An anomaly is an observation, data state, or behavior that substantially deviates from an expected or prevailing pattern or condition.
Outlier Detection in Python
An extreme value is an observation far from the typical level of a time series, either unusually small or unusually large.
- How it is used
- Identifies unusually high or low time-series observations for investigation, including sales counts and industrial sensor readings.
- How deep it goes
- Works through IQR-based limits and threshold tuning, then contrasts fixed rules with forecast-based detection rather than deriving forecasting models.
- Examples
- Uses an injected daily sales spike, plotted against IQR limits, and industrial sensor deviations.
- What you can do after
- Apply it via Python IQR limits to flag candidate extreme values, then investigate what the flags mean.
Statistics Slam Dunk
An outlier is an observation that lies unusually far from the other observations in a distribution or dataset.
- How it is used
- Protects linear regression analyses from extreme observations that can bias model results.
- How deep it goes
- Explains visual detection, two statistical tests, and winsorization, while noting limitations of Dixon’s Q Test and Grubbs’ Test.
- Examples
- Uses NBA career win shares and player deflections, comparing scatterplots, histograms, and boxplots before and after capping values.
- What you can do after
- Diagnose/debug with it by visually identifying extreme values and capping them through winsorization before regression modeling.
The Well-Grounded Data Analyst
An outlier is an observation whose value differs substantially from the typical pattern of the other observations.
- How it is used
- Uses outliers to guide data-quality checks and uncover meaningful groups before downstream analysis.
- How deep it goes
- Demonstrates visual and record-level inspection plus a price-filtering decision, without defining a formal detection rule.
- Examples
- Uses daily e-commerce event counts, house-sale prices, and high-spending customer groups as examples.
- What you can do after
- Evaluate/understand unusual observations through visualization, record inspection, contextual hypotheses, and explicit cleaning choices.
Machine Learning for Tabular Data: XGBoost, Deep Learning, and AI
An extreme value is an observation or value located unusually far toward an edge of an observed distribution.
- How it is used
- Use outlier checks during exploratory analysis to find data problems and limit values that may affect machine-learning performance.
- How deep it goes
- Explains detection through descriptive statistics, charts, and percentiles, then demonstrates winsorization and percentile-based filtering rather than formal outlier tests.
- Examples
- Uses Auto MPG numeric features, accommodation predictors capped by winsorization, and an accommodation-price target filtered with percentile bounds.
- What you can do after
- Evaluate/understand whether extreme observations are errors, legitimate cases, or candidates for capping or percentile-based exclusion.
Read how Machine Learning for Tabular Data: XGBoost, Deep Learning, and AI explains anomaly →
Data Without Labels
In DBSCAN, an outlier or noise point is an observation that is neither a core point nor reachable from a core point and therefore receives no cluster assignment.
- How it is used
- Uses anomalies to distinguish DBSCAN noise from clustered observations and to flag unusual values that may affect analysis.
- How deep it goes
- Defines noise through DBSCAN neighborhoods and gives practical detection and treatment options, without developing a formal anomaly-scoring method.
- Examples
- Uses transaction-count and rainfall extremes, a core-border-noise diagram, and Python cluster and nearest-neighbor plots.
- What you can do after
- Apply it via scikit-learn DBSCAN to inspect noise points, then evaluate whether unusual observations warrant investigation or treatment.
Starting Data Analytics with Generative AI and Python
An outlier is an observation that is unusually distant from the other values in a dataset or from an expected pattern.
- How it is used
- Supports data-quality checks and interpretation of unusual observations in business analysis.
- How deep it goes
- Defines outliers and outlines detection, investigation, and treatment choices without prescribing a universal decision rule.
- Examples
- Uses transaction locations and review response-time distributions to illustrate unusual observations.
- What you can do after
- Evaluate unusual observations using counts and visualizations before choosing a context-appropriate treatment.
Read how Starting Data Analytics with Generative AI and Python explains anomaly →
Cyber Threat Hunting
An outlier is an observation that lies unusually far from the central or typical range of a data set.
- How it is used
- Uses outlier handling to reduce statistical noise while investigating anomalous network behavior and possible beaconing.
- How deep it goes
- Explains quartiles, skew, and IQR filtering with pandas listings, then stops short of fully interpreting each candidate connection.
- Examples
- Demonstrates outlier filtering on network timing data grouped by source IP, destination IP, and destination port.
- What you can do after
- Diagnose noisy network timing data with pandas and produce lower-variance connection candidates for threat investigation.
Pandas Workout: 200 exercises to make you a stronger data analyst
An outlier is an observation unusually far below or above the central range of a dataset according to a chosen criterion.
- How it is used
- Uses unusual observations to investigate precise analytical questions and assess whether changes affect rare or extreme cases.
- How deep it goes
- Explains IQR thresholds, pandas filtering, comparative analysis, boxplots, and z-score extensions without developing statistical theory.
- Examples
- Uses taxi distances and passenger counts, plus Chicago temperature readings visualized with boxplots.
- What you can do after
- Diagnose unusual observations with pandas, compare related measures, visualize extremes, and experiment with percentile or z-score criteria.
Read how Pandas Workout: 200 exercises to make you a stronger data analyst explains anomaly →
Experimentation for Engineers
An observed data point that lies unusually far from the other observations in a dataset or fitted relationship.
- How it is used
- Flags measurements that can make a response-surface regression unreliable near an assumed optimum.
- How deep it goes
- Defines visual diagnosis and range-shrinking remediation, but defers automatic bad-point detection and universal thresholds.
- Examples
- Uses a low markout-profit measurement at threshold 0.5 and order_size 3.5, then contrasts it with a later outlier-free iteration.
- What you can do after
- Diagnose questionable regression measurements and refine parameter ranges before continuing optimization.
Domain-Specific Small Language Models
A model feature or weight with unusually large magnitude that can hurt low-precision quantization if not treated separately.
- How it is used
- Identifies unusually large weights or activations that can undermine low-precision LLM quantization during deployment.
- How deep it goes
- Explains why outliers impair precision and compares threshold-based separation with per-channel smoothing, without deriving the transformations.
- Examples
- Compares FP16, naive 8-bit, and SmoothQuant accuracy results on validation examples, including OPT-2.7B.
- What you can do after
- Evaluate/understand when outlier handling may protect quantized-model accuracy and why accelerator efficiency affects the choice of method.
Read how Domain-Specific Small Language Models explains anomaly →
Build a DeepSeek Model (From Scratch)
An outlier is a numerical activation, weight, or gradient whose magnitude is unusually large relative to other values in its group.
- How it is used
- Motivates replacing tensor-wide scaling with fine-grained scaling to preserve precision during low-precision quantization.
- How deep it goes
- Works through a numerical quantization example and explains grouped scaling as mitigation, without detailing its implementation.
- Examples
- Illustrates precision loss with the vector [2.0, 3.0, 500.0] quantized to 8-bit integers.
- What you can do after
- Understand how a large value can degrade smaller values under shared scaling and how independent group scales mitigate the effect.
Read how Build a DeepSeek Model (From Scratch) explains anomaly →
Statistics Every Programmer Needs
An outlier is an observation that is unusually distant from the pattern represented by the rest of a dataset.
- How it is used
- Uses outlier checks when assessing data and fitting linear regression.
- How deep it goes
- Defines outliers, mentions scatterplot detection, and illustrates removal without detailing a detection procedure.
- Examples
- Removes a presumed point from eight observations and reports R² rising from 0.88 to 0.95.
- What you can do after
- Diagnose regression data by checking for outliers and evaluating their effect on model fit.
Read how Statistics Every Programmer Needs explains anomaly →
Build AI Drug Discovery Pipelines
An outlier is a data point whose value is unusually distant from the pattern or distribution of the other observations.
- How it is used
- Identifies unusual solubility observations that may hinder model learning and introduces low-density regions as possible anomalies.
- How deep it goes
- Defines outliers and gives IQR thresholds, but does not develop a complete anomaly-detection workflow.
- Examples
- Uses aqueous-solubility values from −12 to 1.97 logS and contrasts them with a typical drug-like range near −6 to −3.
- What you can do after
- Diagnose potential outliers by applying IQR thresholds, inspecting chemical extremes, or identifying low-probability regions.
Read how Build AI Drug Discovery Pipelines explains anomaly →
Data-Driven Cybersecurity
An outlier is an observation that lies unusually far from the expected distribution or reference pattern.
- How it is used
- Identifies deviations in cybersecurity data that may warrant threat investigation.
- How deep it goes
- Explains baseline analysis, thresholding, and false-positive review, then demonstrates clustering; it defers deeper statistical and AI analysis.
- Examples
- Uses device communication and login-behavior data to illustrate outliers in clustering plots and outputs.
- What you can do after
- Diagnose potentially unusual device behavior by applying statistical thresholds or inspecting clustering results.
Vibe Engineering
An outlier is an observation that differs markedly from the distribution or expected pattern of the other observations.
- How it is used
- Uses outlier handling in medicine-price prediction to limit extreme or potentially broken prices from influencing model predictions.
- How deep it goes
- Applies a 99th-percentile price cutoff but does not describe how to distinguish valid extremes from broken values.
- Examples
- Uses medicine prices above the 99th percentile as the exclusion case and notes outlier handling as a real-data challenge.
- What you can do after
- Apply a 99th-percentile price filter before fitting a price-prediction model.
The Art of AI Product Development
An outlier is a data point whose behavior or value is anomalous relative to the dominant patterns in a dataset.
- How it is used
- Detects unexpected deviations in time-series behavior and identifies users who do not fit discovered behavioral clusters.
- How deep it goes
- Defines anomalies as unexpected departures and illustrates peaks, troughs, and unmatched users without detailing detection algorithms.
- Examples
- Uses unusual clicking by a new user, a bot outside customer clusters, and a brief traffic spike as examples.
- What you can do after
- Diagnose unusual activity by investigating deviations, as with a user whose repeated clicks revealed a site-scraping bot.
Read how The Art of AI Product Development explains anomaly →
Financial AI in Practice
An outlier is a data value that is unusually distant from other observations in a dataset or distribution.
- How it is used
- Protects credit-model binning and WoE calculations from extreme values that can distort results or destabilize production performance.
- How deep it goes
- Compares capping, transformation, and dedicated bins, but does not provide implementation steps or a method for selecting among them.
- Examples
- Uses unusually high incomes, large transactions, spending spikes, and extremely wealthy borrowers as examples.
- What you can do after
- Apply capping, transformations, or dedicated bins while weighing genuine high-value signals against overfitting to rare observations.
Quantization and Fast Inference
An outlier is a value that differs substantially from most values in a distribution.
- How it is used
- Protects rare, high-magnitude activation dimensions when quantizing large language models.
- How deep it goes
- Explains how outliers disrupt quantization and summarizes mitigation methods without deriving their implementations.
- Examples
- Contrasts activations mostly in [−1,1] with a channel near 100, and describes dimensions hundreds of times larger than neighbors.
- What you can do after
- Evaluate/understand why clipping or coarse scaling can harm quantized model quality and motivate outlier-aware methods.
CUDA for LLMs
An outlier is a data value that differs substantially from typical values in a distribution.
- How it is used
- Uses outliers to motivate robust calibration and finer scale granularity in quantization.
- How deep it goes
- Works through min-max precision loss with numeric values, then contrasts percentile calibration and per-channel or per-group scales.
- Examples
- Uses activations [0.1, 0.5, 0.8, 100.0] to show how an outlier compresses ordinary values under UINT8 asymmetric quantization.
- What you can do after
- Diagnose precision loss from an outlier and compare percentile calibration or finer scale granularity as responses.
Grokking Statistics
An outlier is an observation that differs markedly from most other observations in a dataset.
- How it is used
- Flags unusual observations and explains why median-based summaries can better represent typical values when data is skewed.
- How deep it goes
- Defines outliers through examples and standard-deviation thresholds, stopping short of formal detection methods.
- Examples
- Uses bar tabs, extreme coffee measurements, and an unlikely investment return to illustrate unusual observations.
- What you can do after
- Evaluate/understand how extreme values affect summaries and identify unusually distant observations with standard-deviation ranges.
Also covered in
- Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI
- Microservices Patterns, Second EditionIn early access (MEAP)
- Latency
- Graph Neural Networks in Action
- The Coder CafeIn early access (MEAP)
- Causal AI
- Ensemble Methods for Machine Learning
- Snowflake Data Engineering
- Logs and Telemetry
- Demand Forecasting Best Practices
What to know first
- meanThe mean is a measure of central location equal to the arithmetic average of numerical observations, or the probability-weighted average of a random variable’s possible values.
- standard deviationStandard deviation is the nonnegative square root of variance and measures the dispersion of values around their mean in the same units as those values.
- language modelA language model is a computational model that represents patterns in linguistic sequences to assign probabilities, predict, generate, or evaluate language.
- medianThe median is the middle value in an ordered set of observations, or the average of the two middle values when the set has an even number of observations.
Related concepts
- local outlierincludes · A local outlier is a data item that differs from other items in its nearby or comparable cluster while belonging to a common overall type.
- out-of-distributionincludes · Out-of-distribution describes data generated under conditions or from a distribution that differs materially from those represented in a model’s training data.
- winsorizationaddressed by · Winsorization is a data transformation that replaces extreme observations beyond selected thresholds with less-extreme boundary values rather than removing them.
- box plotused by · A box plot is a statistical graphic that summarizes a numerical distribution using quartiles, a median, and whiskers or markers indicating spread and potential outliers.
- contextual anomalyincludes · A contextual anomaly is an observation that is anomalous relative to its surrounding temporal, situational, or other contextual conditions.
- interquartile rangeuses · The interquartile range is the difference between the third and first quartiles of a distribution, measuring the spread of its middle half.
Questions about anomaly
What is anomaly?
An anomaly is an observation, data state, or behavior that substantially deviates from an expected or prevailing pattern or condition.
Which Manning books explain anomaly?
Outlier Detection in Python, Statistics Slam Dunk, The Well-Grounded Data Analyst, Machine Learning for Tabular Data: XGBoost, Deep Learning, and AI, Data Without Labels, Starting Data Analytics with Generative AI and Python, Cyber Threat Hunting, Pandas Workout: 200 exercises to make you a stronger data analyst and 12 more.
What should I know before learning anomaly?
Manning books that teach anomaly usually introduce mean, standard deviation, language model and median first.
What kinds of anomaly are there?
Kinds of anomaly include local outlier, out-of-distribution and contextual anomaly.
What is anomaly also called?
Anomaly is also called outlier.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.



















