liveBookmanning.com

What you will learn in Data Without Labels

Practical unsupervised machine learning

by Vaibhav Verdhan
Foreword by Ravi Gopalakrishnan

Read it on liveBook Get the book at manning.com

Chapter by chapter

Concepts explained in this book

• clusteringClustering is an unsupervised technique that groups data points so that members of the same group are more similar to one another than to members of other groups.
• dimensionality reductionDimensionality reduction is the transformation of data into fewer dimensions while retaining important structure or information.
• principal component analysisPrincipal component analysis is a linear dimensionality-reduction method that transforms correlated variables into ordered orthogonal components capturing decreasing amounts of variance.
• K-means clusteringK-means clustering is an iterative clustering algorithm that partitions observations into a specified number of groups by minimizing within-group squared distance to cluster centroids.
• hierarchical clusteringHierarchical clustering is a clustering method that constructs a nested hierarchy of groups by successively merging or splitting clusters.
• generative modelingGenerative AI is artificial intelligence that generates new content, such as text, images, or video, from learned patterns and an input prompt or other condition.
• preprocessingData preprocessing, or data preparation, organizes and transforms raw data into a form suitable for analysis or model training.
• data explorationExploratory data analysis is the analytical examination of a dataset to summarize its characteristics, identify patterns, and generate initial insights.
• association rule miningAn association rule is a relationship expressing the co-occurrence or implication of items or events.
• data cleaningData cleaning is the process of detecting and correcting or removing inaccurate, incomplete, inconsistent, duplicate, or irrelevant data.
• singular value decompositionSingular value decomposition is a matrix factorization that expresses a matrix as the product of two orthogonal matrices and a diagonal matrix of singular values.
• distanceA distance metric is a distance function that satisfies the metric-space properties, including nonnegativity, symmetry, and the triangle inequality.
• unsupervised learningAn unsupervised deep learning algorithm trains a multilayer neural model to discover structure or representations without target labels.
• objective functionA loss function quantifies the discrepancy between predicted and actual values for evaluating a model.
• DBSCANDBSCAN is a density-based clustering algorithm that identifies dense regions and labels sufficiently isolated points as noise.
• autoencoderAn autoencoder is a neural-network model that learns to encode data into a lower-dimensional representation and reconstruct the original input from it.
• supportSupport is the proportion of transactions in an association-rule dataset that contain an itemset or all items in a rule.
• language modelA large language model is a machine-learning model trained on very large text collections to understand and generate language.
• t-SNEt-distributed stochastic neighbor embedding is a nonlinear dimensionality-reduction technique that maps high-dimensional observations to a low-dimensional space while preserving local similarities.
• multidimensional scalingMultidimensional scaling is a dimensionality-reduction method that represents objects in a low-dimensional space while preserving specified relationships among their pairwise distances or dissimilarities.
• anomalyIn 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.
• deep learningA deep learning model is a machine-learning model composed of multiple layers of learned representations.
• categorical dataA categorical variable records membership in one of a set of qualitative categories rather than a numeric magnitude.
• Euclidean distanceEuclidean distance is the straight-line distance between points in a coordinate space, computed as the square root of the summed squared coordinate differences.
• labelThe target variable is the response or output value that a supervised model is trained to predict.
• weightA weight is a learned numerical parameter that scales the contribution of an input or intermediate activation in a neural network.
• eigenvalueAn eigenvalue is a scalar associated with a linear transformation and an eigenvector whose direction remains unchanged by that transformation.
• eigenvectorAn eigenvector is a nonzero vector whose direction is unchanged when a linear transformation is applied, changing only by a scalar factor.
• classifierClassification is a predictive modeling task that assigns observations to categorical outcomes or classes.
• centroidA centroid is the arithmetic mean position of a set of points and represents a cluster in centroid-based clustering.
• hyperparameterA hyperparameter is a configuration value selected before or during model training that controls an algorithm's behavior or structure.
• business problemA business problem is a defined organizational objective or challenge that analysis or a data science project seeks to address.
• tokenizationTokenization is the process of splitting text into discrete units such as words or punctuation marks called tokens.
• supervised learningSupervised learning is a machine-learning paradigm in which models learn from input data paired with desired outputs to predict outputs for unseen data.
• activation functionAn activation function is a mathematical function applied to a neuron's combined input to determine its output, typically introducing nonlinearity.
• datasetA collection of related data instances or observations organized for analysis or computation.
• textText data is information represented as written language, including documents, messages, survey responses, and other textual content.
• unstructured dataUnstructured data lacks an easily usable row-and-column schema and may include text, images, audio, or video.
• machine learningAn algorithm that enables a computer system to learn patterns or behavior from data or interaction rather than being explicitly programmed for every outcome.
• domain knowledgeDomain knowledge is subject-matter understanding about the real-world context represented by data and analytical results.
• spectral clusteringSpectral clustering is a clustering method that partitions data using eigenvectors of a graph-based affinity representation rather than only geometric location.
• confidenceConfidence is the conditional proportion of transactions containing the consequent among those containing the antecedent of an association rule.
• liftLift is the ratio of an association rule's observed joint occurrence to the joint occurrence expected if its antecedent and consequent were independent.
• data qualityData quality is the degree to which data is complete, valid, accurate, representative, available, consistent, timely, and internally coherent for a particular use case.
• linear regressionA linear transformation is a mapping that preserves vector addition and scalar multiplication.
• sigmoidA sigmoid function is a differentiable S-shaped function that maps real-valued inputs to bounded output values and is commonly used to produce binary-classification scores.
• lemmatizationLemmatization is a linguistic normalization method that reduces a word to its dictionary or canonical form.
• graph LaplacianA Laplacian matrix is a graph matrix formed in its simplest form by subtracting the adjacency matrix from the degree matrix.
• weighted sumA weighted sum is the sum obtained by multiplying each input by its corresponding coefficient and adding the results.
• degree matrixA degree matrix is a diagonal graph matrix whose entries record the number or total weight of connections incident to each node.
• local neighborhoodA neighborhood is the region around a point containing points at a specified distance from that point, represented as an n-sphere in n-dimensional space.
• Gaussian mixture modelA probabilistic model that represents data as a weighted combination of Gaussian probability distributions.
• locally optimal solutionA local minimum is a point whose objective value is no greater than that of sufficiently nearby points, even if lower values exist elsewhere.
• neuronA neuron is a computational unit in a neural network that combines inputs and applies an activation function to produce an output.
• frequencyFrequency is the number of times an item such as a word or token occurs in a specified text or collection.
• data discoveryData discovery is the process of identifying relevant data sources, assessing access and completeness, and evaluating their feasibility for a project.
• gradient descentGradient descent is an iterative optimization technique that updates parameters in the direction of decreasing objective-function value.
• centroid-based algorithmCentroid-based clustering is a clustering approach that represents each group by a central point and assigns observations according to proximity to those centers.
• backpropagationBackpropagation is an algorithm that computes parameter gradients by propagating error signals backward through a neural network.
• training datasetTraining data is data provided to a machine-learning algorithm during model development so it can learn patterns or parameters.
• sequence miningSequential pattern mining is the discovery of frequent or otherwise interesting ordered patterns from sequences of events or items.
• StandardScalerStandardScaler is a scikit-learn transformer that standardizes features by removing their means and scaling them to unit variance.
• elbow methodThe elbow method is a heuristic for selecting a model complexity value by locating a sharp bend in a curve of objective performance versus that value.
• sentiment analysisSentiment classification is the assignment of text to categories representing expressed opinions, attitudes, or emotional polarity.
• one-hot encodingOne-hot encoding represents each word or category with a binary vector containing one active position and zeros elsewhere.
• thresholdA threshold is a cutoff value that determines whether a neuron's combined input produces a firing or nonzero output.
• convictionConviction is an association-rule metric comparing the expected frequency of antecedent occurrence without the consequent under independence with its observed frequency.
• stressStress is an MDS discrepancy measure that quantifies the difference between original dissimilarities and distances in the fitted low-dimensional representation.
• fuzzy clusteringFuzzy clustering assigns observations degrees of membership to multiple clusters rather than placing each observation in exactly one cluster.
• principal componentA principal component is an orthogonal direction produced by principal component analysis that captures a ranked amount of variation in the data.
• overfittingOverfitting is a modeling condition in which a supervised learning model performs well on training data but less accurately on unseen data.
• market basket analysisMarket basket analysis is a data-mining technique for identifying relationships among items purchased in the same transaction.
• curse of dimensionalityThe curse of dimensionality is the set of difficulties that arise when analyzing data in spaces with many dimensions, including degraded distance behavior, sparsity, increased computation, and overfitting.
• dendrogramA dendrogram is a tree diagram that represents the nested grouping and merging of observations in hierarchical clustering.
• model developmentA model development process is an organized sequence of activities for defining a problem, preparing data, developing a model, and preparing it for deployment.
• deploymentModel deployment is the process of making a trained machine-learning model available to process inputs and produce predictions in an operational environment.
• global optimumA global minimum is a point whose objective value is no greater than that of every point in the domain.
• decision treeA tree-structured predictive model that recursively splits data into nodes using decision rules.
• class probability distributionA probability score is a numerical estimate between zero and one representing the predicted probability of a classification event.
• logistic regressionA classification method that models class-event probabilities using a logistic or sigmoid transformation of input variables.
• similarity matrixAn affinity matrix is a matrix whose entries quantify pairwise similarities or affinities between data points.
• SPADESPADE is a sequential-pattern-mining algorithm that discovers frequent event sequences using a vertical database representation and equivalence-class joins.
• synthetic dataSynthetic data is artificially generated data designed to reproduce selected characteristics or distributions of real data.
• bottleneck layerA bottleneck is the constrained intermediate representation in an autoencoder that retains the most important information from the input.
• border pointA border point is a non-core observation in a DBSCAN cluster that lies within epsilon distance of a core point.
• core pointA core point is a DBSCAN observation having at least minPts points, including itself, within epsilon distance.
• minpointsminPts is the minimum number of points required within a DBSCAN neighborhood to qualify as a dense region.
• reachabilityDensity reachability is the DBSCAN property by which points are connected through a sequence of core-point neighborhood relationships.
• expectation maximizationThe expectation-maximization algorithm is an iterative statistical method that alternates estimating latent-variable assignments and optimizing model parameters.
• Apriori algorithmApriori is an association-rule-mining algorithm that discovers frequent itemsets by generating candidates level by level and pruning them using the downward-closure property of support.
• convolutional neural networkA convolutional neural network is a neural-network model that uses convolutional operations to learn spatial or locally structured features.
• nonmetric multidimensional scalingNonmetric multidimensional scaling is a dimensionality-reduction method that represents ranked dissimilarities in a lower-dimensional space while preserving their order.
• covariance matrixA covariance matrix is a square matrix whose entries represent pairwise covariances among variables.
• orthogonalityOrthogonality is the relationship between vectors or directions whose inner product is zero, corresponding to a right angle in Euclidean space.
• varianceVariance is a measure of the average squared deviation of values from their mean.
• adjacency matrixAn adjacency matrix is a matrix representing which pairs of nodes in a graph are connected and, when applicable, the connection weights.
• embeddingA word embedding maps each word to a numerical vector whose geometry reflects usage-based meaning and relationships.
• Hughes phenomenonThe Hughes phenomenon is the pattern in which a supervised classifier's performance improves as variables are added up to an optimum and then declines when additional variables are included with fixed training data.
• soft clusteringSoft clustering permits an object to belong to multiple clusters or to have graded membership across clusters.
• sequenceA sequence is an ordered collection of events, where the order represents their relative occurrence positions.
• UMAPUniform Manifold Approximation and Projection (UMAP) is a nonlinear manifold-learning technique for dimensionality reduction that constructs a low-dimensional embedding while preserving data relationships.
• membership gradeA membership degree is a value expressing how strongly an observation belongs to a particular fuzzy cluster.
• neural network layerA neural network layer is a computational stage that transforms an input representation into an output representation through connected units and parameters.
• discriminative modelA discriminative model predicts or classifies outcomes by modeling distinctions between data classes or conditional outcomes rather than generating new data.
• interceptA bias term is a learnable additive neural-network parameter that shifts a unit's pre-activation independently of its weighted inputs.
• gradientA gradient is a vector of partial derivatives that indicates how a scalar function changes with respect to multiple variables.
• retrieval rankingRank is an ordered position assigned to an observation or value relative to others.
• stemmingStemming is the heuristic reduction of inflected or derived words to a simplified stem that may not be a valid word.
• deep belief networkA deep belief network is a generative neural-network model composed of multiple layers of stochastic latent variables, typically structured as restricted Boltzmann machines.
• accuracyModel accuracy is the degree to which a predictive model produces correct outputs for the cases being evaluated.
• bag-of-words modelBag-of-words is a text representation that uses word occurrences or frequencies as features while disregarding word order and context.
• Manhattan distanceManhattan distance is the sum of absolute coordinate differences between two points, corresponding to movement along perpendicular axes.
• fuzzy c-means clusteringFuzzy c-means clustering is a fuzzy clustering algorithm that assigns observations degrees of membership across multiple clusters by iteratively updating memberships and cluster centers.
• similarity metricA similarity measure is a quantitative function that assesses how alike two data points, objects, or representations are.
• structured dataA structured dataset organizes values according to a predefined schema, commonly in rows and columns.
• PythonPython is a general-purpose, high-level programming language widely used for software development, data analysis, and machine learning.
• FP-growth algorithmFP-growth is an association-rule-mining algorithm that compresses transactions into an FP-tree and mines frequent itemsets without explicitly generating candidate itemsets.
• frequent pattern treeA frequent pattern tree is a prefix-tree representation of transaction data in which shared item sequences are merged and node counts record item frequencies.
• data science projectA structured project that transforms data into analyses, models, reports, or insights for a defined business problem.
• dataData are recorded facts, measurements, observations, or other representations of phenomena that can be analyzed.
• iris datasetThe Iris dataset is a labeled collection of flower measurements covering three iris species, with 50 samples per species.
• Boltzmann learning ruleThe Boltzmann learning rule is a learning procedure that updates connection weights to reduce the energy of a stochastic neural network while matching data and model distributions.
• fiedler vectorA Fiedler vector is an eigenvector associated with the second-smallest eigenvalue of a graph Laplacian and is used to reveal graph partitions.
• operational dataA transaction is one recorded occurrence containing a set of items or events analyzed together for association relationships.
• probability distributionA probability distribution specifies the probabilities assigned to possible values or configurations of a random system.
• biasA systematic preference or prejudice reflecting unequal social attitudes or structures that can appear in data or model outputs.
• GloVeGloVe is an unsupervised word-representation model that learns vectors from global word co-occurrence frequencies.
• Word2VecWord2Vec is a family of neural language models that learns vector representations of words from their surrounding context in a text corpus.
• hidden layerA hidden layer is an intermediate neural-network layer whose activations are not directly observed as inputs or outputs.
• Cauchy distributionThe Cauchy distribution is a continuous probability distribution with a sharp central peak, slowly decaying heavy tails, and no well-defined mean.
• Kullback–Leibler divergenceKullback-Leibler divergence is a directed measure of the difference between two probability distributions based on their relative information content.
• hard clusteringHard clustering assigns each object decisively to a single cluster.
• data analysisThe systematic examination of data to generate insights, identify trends, and support decisions.
• information lossInformation loss is the loss of meaningful content or distinctions when data is transformed, compressed, or variables are removed.
• data frameA dataframe is a two-dimensional tabular data structure whose rows represent observations and whose columns represent labeled variables that may have different types.
• model interpretationModel interpretation is the analysis of a model's outputs to understand their meaning, validity, and practical usefulness.
• planogramA planogram is a visual plan specifying the placement and arrangement of products within retail space.
• Kaiser criterionThe Kaiser criterion selects principal components whose eigenvalues exceed one.
• scree plotA scree plot is a graph of component eigenvalues or explained variance used to inspect how much information successive components capture.
• artificial intelligenceThe field of computing concerned with creating systems that perform tasks associated with human intelligence, such as reasoning, perception, language use, and learning.
• demonstrationA single historical data record containing input variables and its corresponding target value.
• log-likelihoodA log-likelihood function is the logarithm of the likelihood of observed data as a function of model parameters.
• completenessCompleteness is a data-quality attribute indicating that required values, records, variables, and time periods are present.
• data integrityData integrity is the preservation of accurate and consistent relationships among related tables, variables, and entities.
• data representativenessRepresentativeness is the degree to which data captures the relevant population, business need, and variation without systematic bias.
• validityValidity is the degree to which data conforms to the properties, characteristics, scope, and variations required by its use case.
• uptimeAvailability is the degree to which required data can be obtained and accessed for a particular use case without operational or permission barriers.
• generative adversarial networkA generative adversarial network is a generative model consisting of a generator and discriminator trained adversarially to produce realistic samples.
• contrastive divergenceContrastive divergence is a stochastic approximation technique that estimates the gradient of a generative model's log-likelihood using short Gibbs-sampling chains.
• conditional frequent pattern treeA conditional frequent pattern tree is a frequent pattern tree constructed from the item prefixes that satisfy a specified item's support threshold.
• conditional pattern baseA conditional pattern base is the collection of prefix paths and associated counts leading to a specified item in a frequent pattern tree.
• pretrainingPretraining is an initial model-training stage that learns general patterns or representations before adaptation to a specific task.
• ReLUA rectified linear unit (ReLU) is an activation function that returns its positive input and zero for nonpositive input.
• softmaxA softmax function converts multiple real-valued outputs into nonnegative values that sum to one, typically representing class probabilities.
• normalizationNormalization is the transformation of data values into a common representation or scale so variables can be compared.
• input layerAn input layer is the first neural-network layer that receives input data and distributes it to subsequent layers.
• output layerAn output layer is the final neural-network layer that transforms preceding activations into a prediction or other task output.
• frequent sequenceA sequence of items or events that occurs at least as often as a specified frequency threshold in a dataset.
• support thresholdA support threshold is the minimum frequency or proportion required for a sequence pattern to be considered frequent.
• vertical data formatA vertical data format represents sequence data by storing occurrences with their sequence identifiers and element positions rather than storing complete sequences horizontally.
• term frequency-inverse document frequencyTerm frequency-inverse document frequency (TF-IDF) assigns a term a weight based on its frequency in a document and its inverse frequency across a document collection.
• partition matrixA partition matrix is a matrix whose entries specify the membership degree of each observation in each cluster.
• vanishing gradient problemThe vanishing gradient problem is an optimization difficulty in which gradients become extremely small during learning, hindering the retention of information across many sequence steps.
• ECLATECLAT is a frequent-itemset-mining algorithm that uses a vertical transaction representation and set intersections to compute itemset support.
• speech recognitionSpeech recognition is the computational conversion or identification of spoken language from an audio signal.
• singular valueA singular value is a nonnegative diagonal value in a singular value decomposition that quantifies the magnitude of a corresponding matrix direction.
• singular vectorA singular vector is a vector in a singular value decomposition associated with a left or right orthogonal factor of the decomposed matrix.
• data engineeringData engineering is the discipline of designing and operating systems and processes that collect, transform, store, and deliver data.
• polaritySentiment polarity is a numerical orientation indicating whether text expresses negative, positive, or neutral sentiment.
• subjectivitySubjectivity is a numerical estimate of how much a text expresses opinion rather than factual information.
• TextBlobTextBlob is a Python library for processing textual data, including sentiment analysis and linguistic annotation.
• recurrent neural networkA recurrent neural network is a neural-network architecture designed to model sequences through recurrent state or connections.
• standardizationStandardization is the transformation of numeric variables to a common scale, typically by centering them and scaling by their variability.
• linkage criterionA linkage criterion is a rule that defines the distance between clusters for hierarchical clustering.
• restricted Boltzmann machineA restricted Boltzmann machine is a generative stochastic neural-network model with visible and hidden units connected only across two adjacent layers.
• correlation matrixA correlation matrix is a square table whose entries quantify pairwise statistical association between variables.
• binary dataBinary data has exactly two mutually exclusive possible categories or values.
• continuous dataContinuous data can take any value within a range, including fractional or decimal values.
• contractive autoencoderA contractive autoencoder is an autoencoder regularized to make its representation insensitive to small changes in the input.
• denoising autoencoderA denoising autoencoder is an autoencoder trained to reconstruct clean input data from corrupted or noisy input.
• discrete variableDiscrete data consists of distinct countable values, typically represented as integers.
• nominal dataNominal data consists of categories with no inherent ordering or sequence.
• quantitative dataQuantitative data consists of values that can be measured, weighed, scaled, recorded, and subjected to mathematical operations.
• ordinal dataOrdinal data consists of categories that have an inherent order but not necessarily meaningful numeric spacing.
• qualitative dataQualitative data consists of nonnumeric observations or categories that are not measured or weighed directly.
• sparse autoencoderA sparse autoencoder is an autoencoder trained to produce representations in which most encoded units are inactive or near zero.
• undercomplete autoencoderAn undercomplete autoencoder is an autoencoder whose bottleneck representation has fewer dimensions than its input.
• variational autoencoderA variational autoencoder is a neural-network generative model that learns a probabilistic latent representation and reconstructs data from sampled latent variables.
• decoderA decoder is a neural-network component that transforms an encoded representation back toward the original input form.
• encoderAn encoder is a model component that maps an input into a lower-dimensional or otherwise transformed representation.
• machine translationMachine translation is the automated conversion of text or speech from one human language to another.
• average transaction valueAverage transaction value is total revenue divided by the number of invoices over a specified set of transactions.
• within-cluster sum of squaresWithin-cluster sum of squares measures the summed squared distances of observations from the centroids of their assigned clusters.
• arrayA matrix is a rectangular arrangement of values organized into rows and columns.
• learning rateA learning rate is a scalar hyperparameter that controls the size of parameter updates during iterative training.
• representation learningFeature learning is the automated discovery of useful data representations or features by a model during training.
• similarity graphA similarity graph is a graph whose nodes represent data points and whose edges encode similarity between them.
• fine-tuningFine-tuning is the adaptation of a pretrained model's parameters to optimize performance on a target task.
• metric multidimensional scalingMetric multidimensional scaling is a form of multidimensional scaling that seeks a configuration whose distances preserve the numerical magnitudes of the input dissimilarities.
• reinforcement learningA machine-learning paradigm in which an agent improves its behavior through sequential interactions, trial and error, and feedback from rewards or penalties.
• frequency-based word removalFrequency-based word removal filters text words according to how often they occur in a corpus or dataset.
• lexicon normalizationLexicon normalization maps variant word forms to a common root or canonical form while preserving their shared lexical meaning.
• stop-word removalStop word removal is the elimination of very common words that carry relatively little task-relevant meaning from text.
• generatorA generator is the component of a generative adversarial network that maps input noise or signals to synthetic data.
• pattern recognitionPattern identification is the process of discovering recurring structure or regularity in data.
• versioningVersion control is the systematic management of changes to files, especially source code, so revisions can be tracked, compared, and recovered.
• privacyThe protection of individuals' or organizations' information from unauthorized access, use, or disclosure.
• vocabularyA vocabulary is the set of distinct words or tokens recognized in a corpus or text-processing system.
• anomaly detectionAnomaly detection is the identification of observations or behaviors that deviate substantially from an expected pattern.
• errorPrediction error is the discrepancy between a model's predicted output and the corresponding actual value.
• optimizationOptimization is the systematic selection of values, actions, or configurations to maximize or minimize a defined objective.
• dimensionA dimension is an independent feature or coordinate axis describing each observation in a data representation.
• covarianceCovariance is a measure of how two variables vary together.
• expectation stepThe expectation step of expectation-maximization estimates the probability that each observation belongs to each latent component using current model parameters.
• latent variableA latent variable is an unobserved variable inferred from observed data and a statistical model.
• maximization stepThe maximization step of expectation-maximization updates model parameters using the latent-variable estimates from the expectation step.
• soft assignmentA soft assignment assigns an observation probabilistic degrees of membership across alternative clusters or model components.
• correlationCorrelation is a statistical association describing how changes in one variable relate to changes in another variable.
• discriminatorThe classification component of a generative adversarial network that distinguishes real data from data produced by the generator.

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.