Skip to content

Concept Glossary: D

This page lists English glossary entries for this letter. Entry bodies are assembled from term source files.

data

  • Meaning: Data is information used in different roles: training examples for learning, evaluation material for checking, retrieved evidence for services, user input, state, or operational records.
  • Why it matters: The same word data can mean different things in model training and deployed services. Readers need to distinguish what the information is being used for before judging quality, responsibility, or system behavior.
  • Related concepts: input, output, label, example, state, resource
  • Core Section: P1-4.2
  • Appears in: P1-2.3, P1-3.2, P1-14.1

data leakage

  • Meaning: Data leakage occurs when information unavailable at prediction time enters training or evaluation and makes performance look unfairly good.
  • Why it matters: Leakage breaks the boundary of a fair experiment. A high score may reflect hidden future information or target-like hints rather than real generalization, so data splitting and preprocessing order must be inspected.
  • Related concepts: dataset, validation, target
  • Core Section: P2-12.3
  • Appears in: P4-7.1

data modeling

  • Meaning: Data modeling is the process of turning raw records into samples, features, comparisons, and output structures that can answer a specific question.
  • Why it matters: Many records do not automatically form a learnable problem. Data modeling fixes what counts as one case, what should be compared, and what output form closes the task.
  • Related concepts: dataset, sample, comparison report, output structure, task definition
  • Core Section: P3-1.1
  • Appears in: P3-index, P3-1.2, P3-1.3, P3-2.1, P3-2.2, P3-3.1, P3-3.2, P3-summary

data structure

  • Meaning: A data structure organizes data so particular operations, such as lookup, insertion, traversal, or relationship following, become natural.
  • Why it matters: The right structure depends on which questions are asked often. Lists, dictionaries, trees, and graphs differ because they support different access patterns and costs.
  • Related concepts: abstract data type, linear structure, non-linear structure
  • Core Section: P2-9.1
  • Appears in: P2-9.2, P2-9.3, P2-9.4

DataFrame

  • Meaning: A DataFrame is Pandas' two-dimensional table structure with row and column labels and columns that can hold different types.
  • Why it matters: DataFrames let readers treat data as cases and variables rather than only as numeric arrays. They are the main structure for selecting, filtering, aggregating, and preparing tabular datasets.
  • Related concepts: Series, row, column
  • Core Section: P2-12.1
  • Appears in: P2-12.2, P2-12.3

dataset

  • Meaning: A dataset is a working collection of samples and variables prepared for training or evaluation. It includes choices about inputs, targets, sample boundaries, and splits.
  • Why it matters: The same raw table can become different learning problems depending on how the dataset is built. Dataset design often shapes performance and interpretation before model choice does.
  • Related concepts: DataFrame, feature, target, sample, task definition
  • Core Section: P2-12.3
  • Appears in: P3-2.1, P3-2.2, P3-3.2, P3-9.2

decision tree

  • Meaning: A decision tree is a tree-based model that reaches predictions by asking a sequence of condition questions and following branches to a leaf.
  • Why it matters: Decision trees are intuitive for classification and regression, but deep trees can memorize accidental training patterns. They show the tradeoff between interpretability and overfitting risk.
  • Related concepts: tree, overfitting, split
  • Core Section: P4-14.1
  • Appears in: P4-14.2

alert grade

  • Meaning: An alert grade is a finer operational label for review urgency, such as watch, review, or action. It does not prove that an issue is real; it marks how quickly and carefully the signal should be inspected.
  • Why it matters: A large gap alone should not always trigger immediate action, especially when sample size is small or the signal may be a one-time spike. Alert grades combine difference size, repeatability, recency, and reader impact into a more practical review order. They help operations move from raw signals to prioritized inspection.
  • Related concepts: action bucket, priority, recheck, status inspection, incident record
  • Core Section: P7-7.4
  • Appears in: P7-summary

computation graph

  • Meaning: A computation graph represents operations and value dependencies as connected nodes, making forward and backward calculation paths visible. It breaks a large formula into smaller calculation blocks and their connections.
  • Why it matters: Computation graphs make backpropagation and automatic differentiation easier to understand. They show which intermediate values depend on which earlier operations and how gradient signals can flow backward through the graph.
  • Related concepts: backpropagation, gradient, automatic differentiation
  • Core Section: P5-5.2
  • Appears in: P5-6.1

computational limit

  • Meaning: A computational limit is the point where the candidate space grows too large to inspect every case within realistic time and resources. The issue is not merely that the computer is slow; the search space itself becomes too large.
  • Why it matters: Computational limits explain why complete search can be possible in theory but unusable in practice. They motivate heuristics, pruning, priority strategies, and approximate solutions in real systems.
  • Related concepts: search, search space, exhaustive search
  • Core Section: P1-7.1
  • Appears in: P1-index, P1-6.1, P1-7.2, P1-7.4, P1-8.1

decoder

  • Meaning: A decoder is the component that generates the next token or output sequence from current context and internal representations.
  • Why it matters: Decoders explain output generation in Seq2Seq and GPT-style models. They separate the role of reading input from the role of continuing or producing output step by step.
  • Related concepts: encoder, Encoder-Decoder, GPT
  • Core Section: P1-11.3
  • Appears in: P1-11.2, P6-5.1, P6-20.1

deep copy

  • Meaning: A deep copy creates a new copy of nested internal values as well as the outer container, separating the copy from the original.
  • Why it matters: Deep copies matter when preprocessing or experiments must preserve the original data. They prevent hidden shared inner objects from changing together.
  • Related concepts: shallow copy, reference, list
  • Core Section: P2-8.7
  • Appears in: P2-11.4

deep learning

  • Meaning: Deep learning is a machine-learning approach that uses multi-layer neural networks to learn intermediate representations and prediction rules together.
  • Why it matters: Deep learning marks the move from hand-designed features toward learned representations. It is a specific neural-network-centered approach inside machine learning, not a synonym for all AI.
  • Related concepts: representation learning, parameter, feature
  • Core Section: P1-2.3
  • Appears in: P1-1.3

dependency

  • Meaning: A dependency is an external package or environment condition that code relies on to run.
  • Why it matters: Code files alone are not enough for reproducibility. Dependencies explain why the same code may run in one environment and fail in another.
  • Related concepts: package, reproducibility, requirements.txt
  • Core Section: P2-7.5
  • Appears in: P2-10.1, P2-10.2, P2-10.3

deployment

  • Meaning: Deployment is the act of putting a checked document, site, or service result into the actual public or user-facing state.
  • Why it matters: Deployment changes what users can see or use, so it requires branch discipline, build checks, link checks, and asset checks. File editing and public reflection are different states.
  • Related concepts: branch, static deployment, document reproducibility
  • Core Section: P7-7.1
  • Appears in: P2-14.2, P7-7.2

deployment check

  • Meaning: A deployment check verifies build state, CI status, public URL behavior, and whether the latest content is actually visible after deployment.
  • Why it matters: Static-site deployment can involve several states: local build success, workflow success, Pages update, and reader-visible content. Deployment checks keep those states separate.
  • Related concepts: deployment, status inspection, incident record, review, recency
  • Core Section: P7-7.1
  • Appears in: P7-7.2, P7-7.3

derivative

  • Meaning: A derivative is the instantaneous rate of change: how much output changes when input changes by a very small amount.
  • Why it matters: Derivatives connect function behavior to optimization and learning. They prepare readers to understand gradients and why training follows directions that reduce loss.
  • Related concepts: rate of change, slope, gradient
  • Core Section: P2-4.3
  • Appears in: P2-4.4, P2-6.1, P2-6.3

dictionary

  • Meaning: A dictionary is a mapping data structure that finds values by key rather than by position.
  • Why it matters: Dictionaries help read settings, JSON responses, metadata, and named fields. They separate position-based access from name-based access.
  • Related concepts: key, list, mapping
  • Core Section: P2-8.3
  • Appears in: P2-8.4, P2-9.4

diffusion model

  • Meaning: A diffusion model is a generative model family that starts from a noisy state and gradually denoises it into a meaningful image or representation.
  • Why it matters: Diffusion models show that generation is not always next-token continuation. Image generation often depends on iterative refinement over many steps.
  • Related concepts: generation, next-output generation, sampling
  • Core Section: P1-10.2

dimension

  • Meaning: A dimension is the number of values or coordinate axes used to represent a vector or data point.
  • Why it matters: Dimensions help readers understand vector shape, matrix shape, feature count, and why high-dimensional search or visualization becomes difficult.
  • Related concepts: vector, shape, matrix
  • Core Section: P2-3.2
  • Appears in: P2-3.3, P2-11.1, P2-11.2

dimensionality reduction

  • Meaning: Dimensionality reduction moves high-dimensional data into fewer axes while trying to preserve important structure.
  • Why it matters: It makes complex data easier to visualize or summarize, but the reduced view is not a perfect copy of all original information.
  • Related concepts: unsupervised learning, dimension, representation
  • Core Section: P4-18.1
  • Appears in: P1-8.2, P4-2.2, P4-7.4, P4-18.2

direct lineage

  • Meaning: A direct lineage is the research path that contributes the core structure and problem setting of a current technology.
  • Why it matters: The concept separates background influence from structural ancestry. For LLMs, language modeling, Seq2Seq, Attention, and Transformers matter more directly than every famous deep-learning milestone.
  • Related concepts: surrounding evidence, language modeling, Transformer
  • Core Section: P6-19.2
  • Appears in: P1-9.3, P6-19.1

aggregation

  • Meaning: Aggregation turns many rows or values into smaller summary values such as a mean, sum, count, or ratio. It is a way to read a table through grouped or summarized patterns rather than through each row one by one.
  • Why it matters: Large tables do not answer questions just by being large. Aggregation helps decide which summary represents the current question and whether a pattern appears only after rows are grouped. It also warns readers that a single average may hide counts, ratios, or subgroup differences.
  • Related concepts: groupby, filtering, column
  • Core Section: P2-12.2
  • Appears in: P2-12.3, P3-1.2

directed graph

  • Meaning: A directed graph represents connections with direction, so source and destination are distinct.
  • Why it matters: Direction changes the meaning of relationships such as links, dependencies, follows, or flows. It enables questions about reachability, cycles, and ordering.
  • Related concepts: undirected graph, graph, weight
  • Core Section: P2-9.3
  • Appears in: P2-9.4

array

  • Meaning: An array is a structure that stores values so they can be read by position and index. In numerical computing it usually means a shaped block of values arranged along axes and processed together.
  • Why it matters: Arrays are not only containers; they are computation shapes. Understanding arrays prepares readers for shape, axis, broadcasting, slicing, and vectorized operations. It also separates a simple list of values from data arranged so that whole groups of values can be computed at once.
  • Related concepts: table, index, data structure
  • Core Section: P2-9.2
  • Appears in: P2-11.1, P2-11.2, P2-12.1

batch normalization

  • Meaning: Batch normalization normalizes activation distributions by using the mean and variance of a batch so the next layer receives values in a more manageable range. It is an internal stabilization step, not only input preprocessing.
  • Why it matters: In deep networks, shifting activation distributions can make training unstable. Batch normalization helps stabilize computation and also explains why some layers behave differently in training mode and evaluation mode, where current batch statistics and accumulated statistics are handled differently.
  • Related concepts: training mode, evaluation mode, numerical stability
  • Core Section: P5-8.3
  • Appears in: P5-6.3, P5-8.1

batch

  • Meaning: A batch is a group of items processed together. In operations it can mean grouping many requests or jobs; in deep learning it means a group of samples computed at once. The shared idea is treating multiple items as one processing unit.
  • Why it matters: Batches affect throughput and cost in services, and they are a basic computation unit in deep learning. Batch size changes memory use, gradient noise, and training speed, so a batch is both an operational grouping and a learning-procedure design choice.
  • Related concepts: throughput, operation, tensor
  • Core Section: P1-14.6
  • Appears in: P5-6.1, P5-9.2

discovery

  • Meaning: Discovery is the step of checking what tools, resources, and capabilities a connected system provides before executing actions.
  • Why it matters: Tool use requires knowing what is actually available, what inputs are required, and what permissions apply. Discovery prevents plans from assuming nonexistent capabilities.
  • Related concepts: Model Context Protocol, MCP, tool call, resource
  • Core Section: P1-14.4

distance

  • Meaning: Distance is a numerical measure of how far apart two vectors are in a representation space.
  • Why it matters: Distance turns “close” or “similar” into a computable comparison rule for ranking candidates in search and retrieval systems.
  • Related concepts: similarity, nearest neighbor, vector space, similarity search
  • Core Section: P1-13.2

distributed representation

  • Meaning: A distributed representation expresses a word or object as a vector spread across many numerical dimensions instead of one symbol.
  • Why it matters: Distributed representations support flexible generalization because related items can occupy nearby positions through many partially shared dimensions.
  • Related concepts: embedding, vector, word2vec
  • Core Section: P1-11.1
  • Appears in: P1-9.3, P1-13.1, P5-2.2, P6-3.1, P6-19.2

distribution

  • Meaning: A distribution describes where values concentrate and how they spread across a dataset.
  • Why it matters: Distributions reveal skew, outliers, multimodal structure, and variation that a single average can hide.
  • Related concepts: probability, mean, variance
  • Core Section: P2-5.2
  • Appears in: P2-5.3, P2-13.1, P2-13.2

boolean mask

  • Meaning: A boolean mask selects values by using a true-or-false array of the same length or compatible shape. Instead of naming positions directly, it first evaluates a condition and then keeps the elements where the condition is true.
  • Why it matters: Boolean masks support condition-based selection in preprocessing and array work. They differ from simple slicing because the result shape and copying behavior can change, so readers need to understand them before interpreting filtered samples or conditional array operations.
  • Related concepts: filtering, fancy indexing, copy
  • Core Section: P2-11.4
  • Appears in: P2-12.2

document reproducibility

  • Meaning: Document reproducibility is the ability to rebuild and explain the same document result from the same manuscript, code, images, settings, and procedure.
  • Why it matters: Public documentation depends on more than text. Assets, links, build settings, and deployment steps all affect the final page and must be traceable.
  • Related concepts: reproducibility, deployment, Git
  • Core Section: P2-14.2
  • Appears in: P7-7.1, P7-7.2

documentation

  • Meaning: Documentation records learning content or work process in a structured form that can be reviewed later.
  • Why it matters: Documentation separates feeling that one understood something from being able to explain, verify, and share the reasoning and evidence.
  • Related concepts: personal learning, evidence review, record
  • Core Section: P1-16.1
  • Appears in: P1-16.2, P7-7.1, P7-7.2

dropout

  • Meaning: Dropout is a regularization technique that randomly disables some node outputs or connections during training.
  • Why it matters: Dropout discourages a network from relying too heavily on a few convenient paths. It also explains why training mode and evaluation mode can behave differently.
  • Related concepts: regularization, generalization, training mode
  • Core Section: P5-8.2
  • Appears in: P5-6.4, P5-8.1, P5-8.3, P5-summary

dtype

  • Meaning: dtype records the data type used to store values in a NumPy array.
  • Why it matters: dtype affects memory use and numerical behavior. Arrays with the same visible values can behave differently if they use different integer or floating-point types.
  • Related concepts: ndarray, NumPy, shape
  • Core Section: P2-11.1
  • Appears in: P2-11.2, P2-12.1, P2-12.2