Dataset
A dataset is an organized collection of related data items, records, or observations treated as a unit for storage, processing, analysis, or computation.
LLMs in Production
A dataset is an organized collection of data used for analysis, evaluation, or machine-learning development.
- How it is used
- Supplies data for LLM training, finetuning, evaluation, and retrieval-augmented generation.
- How deep it goes
- Works through extraction, streaming, filtering, tokenization, and splitting, while shortcutting extensive dataset gathering and cleaning in the training example.
- Examples
- Uses Slack messages for a chatbot, TinyStories for streamed training data, and a Python coding dataset for retrieval.
- What you can do after
- Apply it via Python and Hugging Face datasets to extract or stream data and prepare it for training and validation.
Statistics Slam Dunk
A dataset is a collection of structured observations and variables representing recorded information for analysis.
- How it is used
- Supports end-to-end sports analysis by supplying observations to wrangle, explore, visualize, test, and model.
- How deep it goes
- Explains importing, dimensions, records, variables, and extensive wrangling, while deferring dataset design and collection methods.
- Examples
- Uses NBA player-season statistics, franchise histories, box scores, game logs, and play-by-play records.
- What you can do after
- Apply it via R by importing, inspecting, filtering, selecting, converting, reshaping, joining, and summarizing datasets for analysis.
Deep Learning with PyTorch, Second Edition
A collection of examples and annotations used for model training or evaluation.
- How it is used
- Provides indexed examples for model training and validation, connecting raw data to PyTorch's batching pipeline.
- How deep it goes
- Works through the two required methods, a two-class shortcut, custom CT loading, and split safeguards rather than surveying every Dataset variant.
- Examples
- Uses CIFAR-10 bird and airplane images and annotated CT candidates to show both quick filtering and custom item retrieval.
- What you can do after
- Build it from scratch as a PyTorch Dataset that retrieves tensor-ready samples, then keep its training and validation examples separate.
Read how Deep Learning with PyTorch, Second Edition explains dataset →
Grokking Machine Learning
A structured collection of data points intended for analysis or machine-learning tasks.
- How it is used
- Uses datasets to build decision trees and to prepare, train, validate, test, and evaluate machine-learning models.
- How deep it goes
- Explains feature-based splitting with app-download examples and outlines cleaning, encoding, feature selection, and data partitioning.
- Examples
- Uses app-download records grouped by platform or age, then describes numerical and categorical features such as age and animal types.
- What you can do after
- Apply it via Pandas and Scikit-Learn to prepare datasets and train, test, and evaluate machine-learning models.
Designing Deep Learning Systems: A guide for software engineers
A unit of collected data that may be related and is managed as a resource for exploration or model training.
- How it is used
- Positions datasets as the managed data resource for exploration, labeling, transformation, model training, and reproducible experiments.
- How deep it goes
- Explains dataset-manager flows and an internal design using commits, snapshots, metadata, and object-storage links.
- Examples
- Uses public ImageNet in JupyterLab and a storage design with ingested commits, versioned snapshots, metadata, and Amazon S3-style object storage.
- What you can do after
- Evaluate/understand how versioned snapshots and code source control make experimental training inputs reproducible and traceable.
Read how Designing Deep Learning Systems: A guide for software engineers explains dataset →
Machine Learning System Design
A dataset is an organized collection of data records or examples used for analysis, training, evaluation, or other computational purposes.
- How it is used
- Supplies the examples an ML system needs and makes source selection depend on its goal.
- How deep it goes
- Works through source choices and a retail forecasting dataset design without prescribing a fixed preparation sequence.
- Examples
- Uses retail sales forecasting by date, product, and store, alongside a hard-hat detector facing a data cold start.
- What you can do after
- Practice/perform dataset planning by defining prediction units and sources, then checking pipeline reproducibility, consistency, and availability.
D3.js in Action, Third Edition
A dataset is a structured collection of related data values treated as a unit for analysis or visualization.
- How it is used
- Supports D3 data visualization by providing values that can be formatted, measured, bound to graphics, filtered, and interactively updated.
- How deep it goes
- Explains dataset formats, object structure, preparation, binding, scaling, binning, and filtering, while deferring substantial data cleaning.
- Examples
- Uses an employee table, an array of objects, a bar-chart workflow, histogram bins, and gender-filtered histogram states.
- What you can do after
- Builds D3 visualizations by loading, transforming, inspecting, binding, scaling, filtering, and updating dataset-driven graphics.
The Complete Obsolete Guide to Generative AI
A dataset is a structured collection of data used to analyze patterns or train a computational model.
- How it is used
- Supports generative-AI training, LLM-assisted analytics, and sentiment-analysis experiments.
- How deep it goes
- Moves from training-data context through machine-readable analytics and a coded CSV sentiment test, but defers fine-tuning dataset construction.
- Examples
- Uses a call transcript, a business spreadsheet, and 1,000 prelabeled X product-review messages.
- What you can do after
- Apply it via GPT APIs to query data and compare generated sentiment labels with existing labels.
Read how The Complete Obsolete Guide to Generative AI explains dataset →
Software Testing with Generative AI
A dataset is a collection of related data items organized for analysis, processing, or use by a system.
- How it is used
- Supports test-data creation and the preparation of formatted examples for training or fine-tuning language models.
- How deep it goes
- Defines dataset scope and demonstrates structured generation, quality concerns, and synthetic-data trade-offs without detailing storage or evaluation internals.
- Examples
- Uses JSON room records, delimited field specifications, code examples, and question-and-response collections.
- What you can do after
- Apply it via structured prompts to create training or test records, then verify their values, relationships, and relevance.
Read how Software Testing with Generative AI explains dataset →
Data Without Labels
A collection of related data instances or observations organized for analysis or computation.
- How it is used
- Supplies the business-relevant observations and attributes used for analysis, unsupervised learning, and actionable decisions.
- How deep it goes
- Explains dataset selection, derived attributes, quality checks, and a worked review-data workflow; does not specify a universal dataset schema.
- Examples
- Uses retail and telecom attributes, plus collected customer reviews stored in a DataFrame and inspected through rating counts.
- What you can do after
- Practice/perform dataset selection, access checks, attribute derivation, initial inspection, and refresh planning for a business analysis.
Grokking AI Applications
A dataset is an organized collection of data examples used to train or evaluate a machine-learning model.
- How it is used
- Supplies varied training examples that enable models to learn text patterns and produce contextually appropriate completions.
- How deep it goes
- Explains dataset diversity, content sources, and scale, then connects examples to iterative parameter adjustment.
- Examples
- Uses GPT-3's 45 TB collection of webpages, books, Wikipedia, and academic articles as a concrete example.
- What you can do after
- Understand how dataset breadth supports model performance across different contexts and topics.
Build a Large Language Model (From Scratch)
A collection of examples used for training or experimentation.
- How it is used
- Supplies tokenized training pairs for GPT pretraining and padded labeled examples for classification fine-tuning.
- How deep it goes
- Explains tokenization, windowing, padding, balancing, splitting, Dataset indexing, and DataLoader batching without covering storage systems or distributed pipelines.
- Examples
- Uses The Verdict token stream, an SMS spam dataset, shifted token arrays, padded message batches, and train–validation–test splits.
- What you can do after
- Build it from scratch using tiktoken, PyTorch Dataset, and DataLoader to produce training, validation, and test batches.
Read how Build a Large Language Model (From Scratch) explains dataset →
Grokking AI Algorithms, Second Edition
A dataset is an organized collection of data examples considered together for analysis or machine learning.
- How it is used
- Supports machine learning by organizing examples from which algorithms find patterns, make predictions, or generate outputs.
- How deep it goes
- Defines rows, features, quality, diversity, curation, and licensing, but stops short of detailed dataset-processing algorithms.
- Examples
- Uses diamond measurements, agricultural sensors, apartment prices, news archives, academic papers, and image-caption pairs.
- What you can do after
- Evaluate a dataset’s structure, relevance, quality, diversity, representation, and permitted use before applying it to model training.
Read how Grokking AI Algorithms, Second Edition explains dataset →
Cyber Threat Hunting
A dataset is a structured collection of related observations or records for analysis or model training.
- How it is used
- Supports beaconing investigations and machine-learning workflows by organizing events or labeled DNS queries for analysis.
- How deep it goes
- Walks through grouping, filtering, type conversion, and derived-feature calculation, then states training-data quality criteria.
- Examples
- Uses network connection intervals, a 320,044-row event DataFrame, Dropbox connection exercises, and labeled DNS tunneling queries.
- What you can do after
- Apply it via pandas to load, filter, transform, and derive features from security-event datasets for analysis or model training.
Machine Learning Platform Engineering
A dataset is a collection of data instances, optionally paired with labels for supervised learning.
- How it is used
- Supports model training, fine-tuning, inference monitoring, and data-drift analysis across production ML workflows.
- How deep it goes
- Explains datasets through training and drift-monitoring workflows, including dataset-level and column-level metrics, but not storage design.
- Examples
- Uses labeled ID-card images, tabular recommendation data, and historical-versus-current datasets for drift reports.
- What you can do after
- Diagnose data drift by comparing reference and current datasets with Evidently reports and dataset-level metrics.
Read how Machine Learning Platform Engineering explains dataset →
AI-Powered Developer: Build software with ChatGPT and Copilot
A dataset is an organized collection of records and attributes representing a particular subject or purpose.
- How it is used
- Generates realistic records for experimenting with an IT asset management database.
- How deep it goes
- Specifies fields, constraints, relationships, row counts, and alternative generation approaches, but does not establish data-quality validation.
- Examples
- Uses ITAM assets, funding details, depreciation strategies, SQL inserts, and Faker-generated Python records.
- What you can do after
- Generate synthetic ITAM records through PostgreSQL SQL or Python tooling after comparing each approach’s suitability.
Read how AI-Powered Developer: Build software with ChatGPT and Copilot explains dataset →
Microsoft Azure in Action
A dataset is a collection of structured examples or records used to analyze data or develop and evaluate machine-learning models.
- How it is used
- Supports model development, evaluation, fairness checks, and traceable deployment within an AI project.
- How deep it goes
- Explains dataset organization, representation, fairness validation, and provenance, but stops short of dataset schema design or detailed preprocessing code.
- Examples
- Uses reviews, reading histories, demographic segments, genres, JSONL files, and Blob Storage ingestion for a book recommendation engine.
- What you can do after
- Apply it via AI Foundry by organizing data, supplying JSONL training files, and checking representation, fairness, and provenance.
Delivering Machine Learning Projects: From design to deployment
A data set is a collection of data instances available for analysis or machine learning.
- How it is used
- Supports ML delivery by being opened, tested, versioned, pipelined, and delivered to the team.
- How deep it goes
- Explores model-ready preparation, pipeline robustness, data fusion, and population or temporal bias.
- Examples
- Uses controlled and observational data, test-negative protocols, operational snapshots, and data collected across different periods.
- What you can do after
- Operate/configure pipelines that deliver tested datasets to ML teams and identify risks when combining sources.
Read how Delivering Machine Learning Projects: From design to deployment explains dataset →
Deep Learning with R, Third Edition
A dataset is a collection of data instances organized for analysis or computational learning.
- How it is used
- Supports model learning and evaluation by supplying curated, prepared, and appropriately partitioned data.
- How deep it goes
- Covers curation checks, preprocessing, splits, normalization, batching, and generated time-series windows, but not dataset storage internals.
- Examples
- Uses curated labeled data, normalized weather measurements, overlapping integer windows, and train, validation, and test dataset objects.
- What you can do after
- Operate/configure datasets for training, validation, testing, normalization, batching, and time-series window generation.
Read how Deep Learning with R, Third Edition explains dataset →
Distributed Machine Learning Patterns
A collection of data records or examples used for machine learning or other computational analysis.
- How it is used
- Supports scalable model training when incoming or historical data exceeds one machine’s storage or computational capacity.
- How deep it goes
- Explains partitioning, sequential consumption, parallel workers, and trade-offs, but does not specify framework-level configuration.
- Examples
- Uses a 30 GB dataset split into three 10 GB partitions and contrasts sequential model transfer with parallel workers.
- What you can do after
- Evaluate partitioning and sequential or parallel consumption strategies for datasets that exceed single-machine capacity.
Read how Distributed Machine Learning Patterns explains dataset →
Generative AI in Action
A dataset is an organized collection of data records or examples assembled for analysis or machine-learning development.
- How it is used
- Supports retrieval, fine-tuning, and evaluation workflows by supplying external knowledge, task examples, or test inputs.
- How deep it goes
- Covers dataset roles and formats, but stops short of detailed schema design, preprocessing, validation, or data-governance procedures.
- Examples
- Uses external RAG data, an EmojiBot fine-tuning set, and CSV or JSONL evaluation data.
- What you can do after
- Operate/configure RAG updates, fine-tuning preparation, and Azure AI Studio evaluation runs with suitable dataset inputs.
Build an AI Agent (From Scratch)
A dataset is a structured collection of representative examples used for analysis or evaluation.
- How it is used
- Uses representative requests to evaluate agent responses, tool actions, errors, and real-world consequences.
- How deep it goes
- Distinguishes benchmark, custom, and production sources, then connects examples to traces, failure analysis, and metrics.
- Examples
- Includes GAIA, SWE-bench, HotpotQA, custom product scenarios, and production user interactions.
- What you can do after
- Build it from scratch by assembling representative scenarios, running agents, reviewing traces, and expanding coverage over time.
Read how Build an AI Agent (From Scratch) explains dataset →
Deep Learning with Python, Third Edition
Collections of examples used to train or evaluate machine learning models, here including large-scale internet-derived media and text corpora.
- How it is used
- Organizes model development into training, validation, and final testing against unseen data.
- How deep it goes
- Explains split purposes, validation overfitting, information leaks, hold-out validation, and k-fold procedures with code.
- Examples
- Demonstrates shuffled hold-out splitting and three-fold cross-validation through executable array operations.
- What you can do after
- Apply hold-out or k-fold validation and reserve an untouched test dataset for final evaluation.
Read how Deep Learning with Python, Third Edition explains dataset →
Data-Driven Cybersecurity
A structured collection of data observations and attributes prepared for analysis or computational use.
- How it is used
- Supports cybersecurity analysis by organizing device activity or failed-login records for anomaly detection and actionable reporting.
- How deep it goes
- Defines datasets through fields and observations, then demonstrates feature extraction, clustering, and reporting.
- Examples
- Uses five-device login activity with bandwidth and access-time attributes, plus failed-login records grouped by user, hour, device, and reason.
- What you can do after
- Apply datasets via pandas, scikit-learn, and Matplotlib to cluster device behavior and report failed-login patterns.
Related concepts
- fine-tuningused by · Fine-tuning is adapting a pretrained machine-learning model by training it further on targeted data or objectives for a task, domain, or behavior.
- language modelused by · A language model is a computational model that represents patterns in linguistic tokens to estimate sequence probabilities and predict or generate language from context.
- training datasetincludes · A collection of examples or observations supplied to a learning procedure to fit or adapt a model or learn its patterns or parameters.
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.