liveBookmanning.com

What you will learn in Machine Learning for Tabular Data: XGBoost, Deep Learning, and AI

XGBoost, Deep Learning, and AI

by Luca Massaron,Mark Ryan

Read it on liveBook Get the book at manning.com

Chapter by chapter

Concepts explained in this book

• deep learningA machine learning approach based on neural networks with multiple learned layers of representation.
• XGBoostA predictive model trained using XGBoost's gradient-boosted decision-tree algorithms.
• gradient boostingA decision-tree ensemble model built by sequentially adding trees to optimize a predictive loss using gradient-based optimization.
• structured dataData organized according to a defined schema that makes its elements consistently interpretable and manageable.
• feature engineeringFeature processing is the transformation or elaboration of input features to make them suitable or more informative for machine learning models.
• missing valueMissing data is the absence of recorded or available values or cases in a dataset, often represented explicitly by a special marker.
• decision treeA decision tree is a predictive model that recursively partitions inputs using feature-based split rules and produces outcomes at terminal leaves.
• data explorationExploration is the investigation of previously untested regions or configurations in a search space.
• feature interactionA feature interaction is a predictive effect produced jointly by two or more input features rather than by each feature independently.
• hyperparameterA hyperparameter is a configuration value selected before or during model training rather than learned directly from the training examples.
• fastaiPyTorch with fastai is a deep-learning software stack combining the PyTorch framework with fastai's high-level modeling abstractions.
• ensembleA technique that combines multiple models or their predictions to improve predictive results.
• regularizationRegularization is a technique that limits model complexity by penalizing parameter values to reduce overfitting and improve generalization.
• L1 regularizationL1 regularization adds a penalty proportional to the sum of absolute model coefficients, encouraging sparse coefficient estimates.
• linear regressionA statistical model that predicts a numeric response as a linear combination of input variables.
• KerasKeras is an open-source deep learning API for Python that provides interfaces for building and training neural-network models.
• categorical dataAn input variable whose values represent membership in discrete categories rather than continuous quantities.
• ColumnTransformerColumnTransformer is a scikit-learn class that applies different transformations to specified subsets of input features and combines their outputs.
• LightGBMLightGBM is an open-source gradient-boosting framework designed for efficient tree-based learning.
• pandaspandas is an open-source Python library providing data structures and operations for manipulating and analyzing structured and tabular data.
• weight decayL2 regularization adds a penalty proportional to the squared values of model parameters to constrain model complexity.
• label encodingOrdinal encoding represents categorical values with integer codes that impose an ordered numerical representation.
• learning rateA learning rate is a hyperparameter that scales the size of successive updates made while fitting a model.
• high-cardinality categorical featureA high-cardinality categorical feature is a categorical feature with many distinct category values relative to the dataset or available computational resources.
• mean absolute errorMean absolute error is the average absolute difference between predicted or imputed values and their reference values.
• early stoppingEarly stopping is an iterative-training termination criterion that halts optimization when validation performance fails to improve for a specified number of iterations.
• pipelineA data pipeline is a sequence of data-processing steps that moves and transforms data between systems or stages.
• DMatrixAn XGBoost data structure optimized for storing labeled training or prediction data and associated metadata.
• Lightning FlashLightning Flash is a high-level deep learning framework built on Lightning for rapid prototyping, baselining, fine-tuning, and application development.
• data leakageInformation leakage, or data leakage, is the use or transfer of predictive information unavailable at prediction time, producing overly optimistic evaluation results.
• traditional machine learningClassical machine learning is the family of non-neural-network machine learning approaches that commonly rely on statistical optimization, similarity comparisons, and engineered features.
• one-hot encodingOne-hot encoding represents each category of a categorical variable as a separate binary feature.
• classifierA classification problem is a prediction task in which the target is a class or label.
• anomalyAn extreme value is an observation or value located unusually far toward an edge of an observed distribution.
• regression coefficientA regression coefficient is a parameter multiplying an input variable that determines its signed contribution to a linear regression model's prediction.
• ordinal dataCategorical data whose values have a meaningful order or rank.
• standardizationA standardized feature is a feature transformed to a normalized scale, commonly with a specified mean and standard deviation.
• correlationBivariate correlation is a statistical measure of association between two variables or features.
• quantitative dataA continuous column contains real-number values on a continuous numeric scale.
• redundant featureA feature that duplicates the values or information of another feature.
• cross-validationCross-validation is a model-evaluation technique that repeatedly partitions data into training and validation subsets.
• log transformationA log transformation applies a logarithm to values, converting multiplicative relationships into additive ones and changing their scale.
• baselineA deliberately simple predictive model used to establish a reference level of performance.
• temporal featureA temporal feature is an input feature derived from dates or time intervals associated with an observation.
• squared lossSquared error is a loss measure computed as the squared difference between a predicted and observed numeric value.
• binary featureA binary feature is an input feature whose value has two possible states, such as indicating whether another value is missing.
• gradientA gradient is a quantity indicating the direction and rate of greatest local increase of an objective with respect to parameters.
• gradient descentGradient descent is an optimization technique that iteratively adjusts parameters in the direction that reduces a differentiable objective function.
• histogramA representation that groups numeric values into bins and records their frequencies or aggregated statistics.
• labelA numeric output variable to be predicted in a regression problem.
• TabNetTabNet is a neural network architecture for tabular data that uses sequential attention to select features.
• scikit-learnScikit-learn is an open-source Python library providing classical machine learning algorithms, preprocessing utilities, model evaluation tools, and workflow components.
• multicollinearityMulticollinearity, or collinearity, is a condition in which predictor features are strongly linearly related.
• estimatorA scikit-learn-compatible object that implements model fitting and typically prediction or transformation operations.
• machine learningA field of computing in which algorithms learn patterns from data to make predictions or decisions.
• SHAPA SHAP value is a feature-attribution value based on Shapley values that quantifies a feature's contribution to an individual prediction.
• k-d treeA k-dimensional tree, or KDTree, is a space-partitioning data structure that organizes points for efficient multidimensional nearest-neighbor queries.
• data frameA pandas DataFrame is a two-dimensional labeled tabular data structure with rows and columns that can hold heterogeneous data types.
• Scikit-Learn pipelinescikit-learn Pipeline is a composite estimator that chains named data transformations and a final estimator into one sequential modeling object.
• biasBias is systematic prediction error arising from restrictive modeling assumptions or insufficient model flexibility.
• information gainSplit gain is the improvement in a tree objective obtained by dividing a node with a candidate feature-based split.
• area under the curveROC AUC is the area under the receiver operating characteristic curve, measuring a binary classifier's ability to rank positive instances above negative instances.
• deep learning stackA deep learning stack is a coordinated combination of a low-level framework, a high-level API, and a library specialized for a deep learning application domain.
• deploymentA production deployment is the operational release of a software system or machine-learning solution for live application use.
• MissingIndicatorMissingIndicator is a scikit-learn transformer that creates binary features indicating missing values.
• missing at randomMissing at random describes a missingness mechanism in which missingness depends on observed data but not on unobserved values after conditioning on the observed data.
• missing completely at randomMissing completely at random describes a missingness mechanism in which missingness is unrelated to observed or unobserved data.
• Gemini for Google CloudGemini for Google Cloud is Google Cloud's generative-AI assistance and development capabilities based on Gemini models.
• random forestAn ensemble of randomized decision trees whose predictions are aggregated.
• lagged featureA lagged value is a time-series feature formed from an earlier observation of a variable.
• Euclidean distanceEuclidean distance is the straight-line distance between two points in Euclidean space.
• geographical clusteringGeographical clustering is the grouping of locations according to spatial proximity or similarity to represent geographic patterns.
• Manhattan distanceManhattan distance is the sum of the absolute coordinate differences between two points.
• low-cardinality categorical featureA low-cardinality categorical feature is a categorical feature with relatively few distinct values or categories.
• interventional approachThe interventional approach computes feature attributions by intervening on feature values according to an assumed causal dependence structure.
• Shapley valueA Shapley value is a cooperative-game-theory allocation that fairly assigns each participant a contribution across all possible coalitions.
• TreeExplainerTreeExplainer is a SHAP explainer for computing feature attributions from tree-based machine-learning models.
• TreeSHAPTreeSHAP is an algorithm for efficiently computing SHAP values for tree-based models.
• violin plotA violin plot visualizes the distribution and density of numeric values, often alongside summary statistics such as the median.
• logitThe logit function maps a probability in (0,1) to its log-odds and is the inverse of the logistic sigmoid.
• metricA quantitative measure used to monitor or assess model performance rather than directly define the training objective.
• cross_validateA scikit-learn function that evaluates an estimator with cross-validation and returns scores and fit or prediction timing results.
• Gini impurityGini impurity is the probability of misclassifying an element labeled according to the class distribution of its subset.
• Google Vertex AIVertex AI is Google Cloud's managed machine-learning platform for developing, training, deploying, and managing machine-learning models.
• read_csvA pandas function that reads delimited tabular data into a DataFrame.
• objective functionA loss function quantifies the discrepancy between model predictions and expected target values for optimization.
• non-iid dataNon-IID data consists of observations that are dependent, drawn from different distributions, or both.
• box-and-whisker plotA graphical summary of a numeric distribution using quartiles, a median, and potential outliers.
• describeThe pandas describe method computes summary descriptive statistics for data columns.
• indicator variableAn indicator variable is a binary variable that records whether a condition such as missingness holds.
• is_naThe pandas isna method identifies values that represent missing data.
• standard deviationStandard deviation measures the typical dispersion of values around their arithmetic mean.
• accuracyAccuracy is the proportion of classification predictions that match the corresponding true labels.
• language modelA large language model is a machine-learning model trained on large text corpora to understand and generate language sequences.
• sparsitySparsity is a property of a data representation in which many entries or coefficients are zero or empty.
• Vertex AI endpointA Vertex AI endpoint is a managed serving interface that accepts requests and returns predictions from a deployed model in Vertex AI.
• data transformationA feature transformation maps input feature values to a different representation, often to simplify relationships for a model.
• baggingA collection of models trained on resampled data whose predictions are aggregated to reduce variance.
• BaggingClassifierBaggingClassifier is a scikit-learn ensemble estimator that fits base classifiers on bootstrap samples and aggregates their predictions.
• mean target encodingA feature encoding technique that replaces each category with a statistic computed from the target values associated with that category.
• pastingPasting is an ensemble technique that trains multiple models on subsamples drawn without replacement and combines their predictions.
• prediction averagingAn averaging ensemble combines predictions from multiple models by averaging their outputs or taking a majority vote.
• cross-sectional dataCross-sectional data contains observations of multiple units measured at or around the same period rather than repeatedly through time.
• grouped cross-validationGroup cross-validation partitions data so that all observations belonging to a group remain in either training or validation folds.
• independent and identically distributedThe IID principle is the assumption that observations are independently sampled and follow the same probability distribution.
• temporal autocorrelationAutocorrelation is correlation between values of a variable and its own values at different positions or times.
• time-series validationTime-based validation evaluates a model using temporally ordered training and validation data that respect the direction of time.
• missing not at randomMissing not at random describes a missingness mechanism in which missingness depends on unobserved values or other unaccounted-for information.
• out-of-core learningOut-of-core learning is an incremental training technique that processes data from external storage in portions without loading the entire dataset into memory.
• logistic regressionLogistic regression is a regression model that estimates class probabilities using a logistic link function.
• discretizationA technique that groups numeric values into discrete ordered bins and assigns each value to a bin.
• fastai tabularfastai tabular is a fastai component providing tabular deep learning models and utilities.
• Bayesian optimizationA Bayesian optimizer is an optimization method that uses outcomes from evaluated configurations to select promising future configurations.
• high-level APIA high-level deep learning API abstracts common neural-network construction and training tasks.
• domain knowledgeDomain knowledge is expertise about the subject matter and phenomena represented by a dataset.
• weak learnerA predictive model that performs only slightly better than a simple baseline and is combined with other models in an ensemble.
• out-of-core computingOut-of-core computing processes datasets too large for main memory by moving data between external storage and memory in chunks.
• colsample_bytreecolsample_bytree is a tree-ensemble hyperparameter specifying the fraction of features considered when splitting each tree.
• random patchesRandom patches is an ensemble sampling technique that trains models on randomly selected subsets of both observations and features.
• entropyEntropy is a measure of disorder or randomness in a distribution of class labels.
• pruningPruning removes selected decision-tree splits after or during construction to reduce tree complexity.
• tree-based methodA tree-based model is a predictive model whose decision structure is formed from one or more decision trees.
• class imbalanceImbalanced data has substantially unequal representation of classes or outcome groups.
• OptunaOptuna is an open-source hyperparameter optimization framework that adaptively searches parameter configurations through repeated trials.
• Hessian matrixThe square matrix of second-order partial derivatives of a scalar multivariable function, describing its local curvature.
• preprocessingData preprocessing is the transformation and preparation of raw data before it is supplied to a machine learning model.
• scikit-learn APIThe scikit-learn API is the library-wide interface convention for fitting, predicting with, transforming data using, and configuring estimators.
• regular expressionA pattern notation for matching, extracting, or transforming text sequences.
• BaggingRegressorBaggingRegressor is a scikit-learn ensemble estimator that fits base regressors on bootstrap samples and aggregates their predictions.
• feature selectionFeature sampling selects a subset of input features as candidates for modeling or a split.
• hyperparameter search spaceA hyperparameter search space is the set of candidate values and constraints from which model configurations are selected during optimization.
• model training scriptA program file containing the code and configuration needed to train a machine-learning model.
• interpretabilityThe degree to which a model's operation and predictions can be understood by people.
• coefficient of determinationR-squared is a goodness-of-fit metric measuring the proportion of variance in the target explained by a predictive model.
• multivariate imputation by chained equationMultivariate imputation replaces missing values by using information from multiple other features or columns.
• recordA record in tabular data that represents one observation or data point.
• correlation matrixA symmetrical matrix containing pairwise correlations among a collection of variables or features.
• Cramer's VA normalized measure of association between categorical variables ranging from 0 to 1 and indicating association intensity without direction.
• dimensionality reductionDimensionality reduction transforms data with many variables into a lower-dimensional representation while attempting to preserve important structure or information.
• principal component analysisA linear dimensionality-reduction method that represents data using orthogonal components formed as weighted combinations of the original features.
• singular value decompositionA matrix factorization that expresses a matrix as the product of orthogonal matrices and a diagonal matrix of singular values.
• t-SNEt-SNE is a nonlinear dimensionality-reduction technique that maps high-dimensional data to a low-dimensional space while emphasizing local neighborhood structure.
• UMAPUMAP is a nonlinear dimensionality-reduction technique based on manifold-learning principles that constructs a low-dimensional representation of high-dimensional data.
• root mean squared errorRoot mean squared error is the square root of the mean squared differences between predicted and observed numeric values.
• category_encoderscategory_encoders is a Python library providing scikit-learn-compatible encoders for categorical variables, including target and ordinal encoders.
• native categorical supportNative categorical support is a model capability for processing categorical features directly without first converting them to numeric codes.
• OneHotEncoderOneHotEncoder is a scikit-learn transformer that converts categorical values into one binary feature per category.
• OrdinalEncoderOrdinalEncoder is a scikit-learn transformer that maps categories to numeric ordinal codes.
• smoothing parameterA smoothing parameter controls the blending of category-specific target statistics with an overall target statistic.
• TargetEncoderTargetEncoder is an encoder implementation that replaces categories with smoothed target-derived statistics.
• PyTorchPyTorch is an open-source machine learning framework for tensor computation and neural-network development.
• gblinear boosterAn XGBoost booster configuration that uses regularized linear models as base learners.
• data augmentationA technique that expands or diversifies training data by applying label-preserving transformations or generating additional examples.
• encoding layerAn encoding layer is a neural-network component that transforms categorical inputs into numeric representations during model computation.
• data warehouseA data warehouse is a centralized system that stores integrated data for analysis and reporting.
• columnar storage engineA columnar storage format stores values from the same table column contiguously to support efficient column-oriented access.
• dimension tableA dimension table stores descriptive attributes that provide context for records in other tables.
• distributed programmingDistributed computing divides computation across multiple networked machines or processors.
• fact tableAn event table stores records of business events measured at particular moments.
• item tableAn item table stores detailed information about products or items associated with business events.
• mergeA merge function combines tabular structures by matching records according to shared keys or join conditions.
• normalized schemaA normalized schema organizes information into separate focused tables to reduce redundancy and improve data integrity.
• SQLSQL is a standardized language for querying and manipulating data in relational databases.
• average precisionA classification metric that summarizes the precision-recall curve, typically by averaging precision over recall increments.
• balanced classification problemA balanced target is a classification target whose classes contain approximately equal numbers of observations.
• binary targetA binary target is a prediction target with exactly two possible classes or values.
• multiclass targetA multiclass target is a categorical prediction target with more than two possible classes.
• target analysisTarget analysis is the examination of a prediction target's values and distribution to assess its suitability for modeling and evaluation.
• polynomial basis expansionPolynomial expansion transforms input features by adding powers and products of features up to a specified degree.
• ridge regressionRidge is a regularized linear regression estimator that penalizes the squared magnitude of coefficients.
• PyTorch LightningPyTorch Lightning is a high-level PyTorch framework that structures and abstracts common model-training workflows.
• TensorFlowTensorFlow is an open-source software platform for building and executing machine learning computations, including neural networks.
• TabularDatasetA managed dataset resource in Vertex AI that organizes data for machine-learning development and training.
• unstructured dataData whose varied modalities and organization do not conform to a fixed rows-and-columns schema.
• Keras preprocessing layerKeras preprocessing layers are neural-network layers that transform input features, such as by normalization or categorical encoding, within a Keras model.
• piecewise linear gradient boosting treeA piecewise linear gradient boosting tree is a boosted tree whose terminal nodes contain linear models for local prediction.
• extremely randomized treesExtremely randomized trees are an ensemble of decision trees that randomly selects the feature for each split and then optimizes the split point on that feature.
• newton-raphson methodA Newton-Raphson update is an optimization step that uses first- and second-order derivatives to refine parameter values.
• FlaskFlask is a lightweight Python web framework for building web applications and HTTP request handlers.
• sigmoidA sigmoid is a smooth mathematical function that maps real-valued inputs to values between zero and one.
• Occam's razorOccam's razor principle favors the hypothesis that explains the evidence with the fewest assumptions among competing hypotheses.
• nonlinear mappingA nonlinear transformation maps inputs to outputs in a way that cannot be represented as a linear function.
• random subspacesRandom subspaces is an ensemble sampling technique that trains each model using a randomly selected subset of features.
• batchingA batch is a subset of training examples processed together during an incremental learning step.
• partial fittingpartial_fit is a scikit-learn estimator method that incrementally trains a model on supplied data, potentially across multiple calls.
• stochastic gradient descentMini-batch gradient descent is an optimization algorithm that updates model parameters using gradients computed from small batches of examples.
• embeddingAn embedding is a learned or constructed representation that maps high-dimensional data into a lower-dimensional space while preserving selected characteristics or relationships.
• individual conditional expectation plotAn individual conditional expectation plot shows the predicted response for an individual observation as a feature varies.
• partial dependence plotA partial dependence plot displays a model's average predicted response as one or more input features vary while marginalizing over other features.
• XGBFIRXGBFIR is a project that extracts and ranks feature interactions from XGBoost models.
• feature matrixA rectangular numerical data structure in which rows represent observations and columns represent input features.
• concept driftConcept drift is a change over time in the relationships between input predictors and the target that can make a previously trained model ineffective.
• histogram aggregationA decision-tree splitting technique that bins continuous feature values and evaluates bin boundaries as candidate split points.
• underfittingUnderfitting is a modeling failure in which a model is too simple to capture important patterns in the data.
• view functionA view function is a Flask request-handler function that processes a web request and returns a response or rendered page.
• scikit-learn transformer classA transformer is a Scikit-learn object that learns a data transformation and applies it to return transformed data.
• StandardScalerStandardScaler is a Scikit-learn transformer that standardizes numeric features using statistics learned from fitting data.
• representation learningRepresentation learning is the automatic discovery of useful data representations from input data.
• cross entropy lossCrossEntropyLoss is a PyTorch loss-function module that combines class-logit normalization with negative log-likelihood for classification.
• TabularDataLoadersTabularDataLoaders is a fastai data-loading abstraction that packages tabular samples, labels, preprocessing, and data splits for model training.
• generative modelingArtificial intelligence that generates new content, such as text, images, code, or structured outputs, from learned patterns.
• pipeline scriptA pipeline script is a program that defines inputs, resources, and execution steps for a machine-learning pipeline.
• normalization layerA normalization layer is a neural-network layer that transforms input feature values to a standardized numeric distribution during model computation.
• modeMajority voting predicts a classification label by selecting the class receiving the most votes from multiple models.
• Vertex AI prebuilt containerA prepackaged software container that provides a configured runtime for executing a machine-learning workload.
• hyperparameter tuningHyperparameter optimization, or tuning, is the process of searching for hyperparameter settings that improve a model's evaluation objective.
• evaluation datasetA validation dataset is a portion of data reserved for evaluating and tuning a model during development rather than fitting its parameters.
• data typeA classification that specifies the kind of values a data field can contain.
• rare categoryA rare class is a target category represented by a small proportion of observations in a dataset.
• Google Dataset SearchA search service that indexes dataset descriptions across online repositories and helps users locate source datasets.
• KaggleAn online platform for data science competitions, datasets, notebooks, and machine learning collaboration.
• kaggle utilityA command-line interface for interacting with Kaggle services, including downloading datasets.
• acquisition functionAn acquisition function scores candidate parameter settings according to their expected value and uncertainty for the next objective evaluation.
• surrogate modelA surrogate function is an approximate model of an objective function that predicts objective behavior across untested parameter settings.
• adversarial validationAdversarial validation is a technique that trains a classifier to distinguish samples from two datasets in order to assess their distributional similarity.
• generative adversarial networkA generative adversarial network is a deep learning architecture consisting of a generator and discriminator trained adversarially to produce samples resembling real data.
• latent spaceA latent space is a lower-dimensional representation space in which a model encodes condensed information about observed data.
• SDVSDV is an open-source software library for generating synthetic data from existing datasets using statistical and machine learning models.
• synthetic dataSynthetic data is artificially generated data designed to resemble the characteristics or distribution of real data without representing actual observations.
• variational autoencoderA variational autoencoder is a neural network architecture that encodes data into a probabilistic latent space and decodes latent representations to reconstruct or generate data.
• SimpleImputerSimpleImputer is a scikit-learn transformer that replaces missing values using a univariate strategy such as the mean or median.
• cross-validation predictionA prediction produced for an observation by a model trained without the validation fold containing that observation.
• generalized linear modelA generalized linear model extends linear regression by combining a linear predictor with a response distribution and link function.
• Vertex AI SDKThe Vertex AI SDK is a software development kit that provides programmatic access to Vertex AI services from supported applications.
• importance scoreFeature importance is a measure of how strongly an input feature contributes to a model's predictions or performance.
• callbackA Keras callback is an object that runs user-defined actions at specified points during model training or evaluation.
• extrapolationExtrapolation is the estimation of values beyond the range of target values observed during training.
• Poisson distributionThe Poisson distribution models the number of events occurring in a fixed interval when events occur independently at a constant average rate.
• Google CloudGoogle Cloud is a cloud-computing platform providing infrastructure and managed services for applications, data, and machine learning.
• descriptive statisticsDescriptive statistics are numerical summaries that characterize the distribution and central, spread, or other properties of observed data.
• KFoldA scikit-learn cross-validation splitter that divides data into k folds for successive training and validation splits.
• bootstrap samplingBootstrap is a resampling technique that creates datasets by sampling observations with replacement from an observed dataset.
• grid searchGrid search is an exhaustive hyperparameter optimization technique that evaluates every combination of specified hyperparameter values.
• SAINTSAINT is a transformer-based neural network architecture for tabular data.
• code compilationModel compilation is the transformation of a trained model into lower-level executable code for faster inference or deployment.
• GPUA GPU is a massively parallel processor designed to execute many computational operations concurrently.
• lleaveslleaves is a tool that compiles LightGBM tree models to LLVM-generated assembly for faster prediction.
• TreeliteTreelite is a tool that compiles decision-tree ensemble models into portable low-dependency C or Java code.
• stability selectionStability selection is a feature-selection technique that evaluates how consistently features are selected across multiple resampled datasets.
• service accountA service account is a non-human identity that applications or automated services use to authenticate and access authorized cloud resources.
• binary treeA tree data structure in which each node has at most two child nodes.
• transfer learningA technique that adapts knowledge learned from one task or dataset to a related task.
• weighted quantile sketchAn approximate quantile algorithm that uses weights, such as gradient statistics, to identify candidate split points.
• CategoryEncodingCategoryEncoding is a Keras preprocessing layer that converts integer category indices into encoded representations.
• StringLookupStringLookup is a Keras preprocessing layer that maps string tokens to integer indices.
• TensorFlow Datasettf.data.Dataset is a TensorFlow data abstraction representing a sequence of elements that can be transformed and iterated during model input processing.
• transparencyThe degree to which a model's operation, predictions, and feature reliance can be understood and assessed by people.
• BorutaBoruta is a feature-selection algorithm that identifies features relevant to a prediction task by comparing them with randomized shadow features.
• Cloud Shell EditorCloud Shell Editor is a browser-based source-code editor integrated with Google Cloud Shell.
• heterogeneityHeterogeneous data contains features with differing semantic types, numeric scales, units, or categorical value structures.
• random rotationA random rotation is a reversible transformation that applies a randomly selected orthogonal rotation to input feature vectors.
• rotation matrixA rotation matrix is a square orthogonal matrix that preserves vector lengths and angles under transformation.
• rotational invarianceRotational invariance is the property of producing equivalent model behavior when input representations are rotated.
• skewA probability or empirical distribution that is asymmetric around its central tendency.
• ChatGPTChatGPT is a conversational AI system based on large language models that generates responses to user prompts.
• f-scoreThe harmonic mean of precision and recall used to evaluate classification performance.
• TableGPTA framework that uses large language models to let users interact with tabular data through natural-language commands.
• TABLET benchmarkA benchmark for evaluating large language models on prediction tasks involving tabular data.
• constant featureA feature whose value is identical for every observation in a dataset.
• exclusive feature bundlingExclusive feature bundling is a feature-dimensionality-reduction technique that groups sparse, mutually exclusive features while preserving their nonzero information.
• gradient-based one-side samplingGradient-based one-side sampling is a sampling algorithm that retains examples with the largest gradients and randomly samples examples with smaller gradients using corrective weights.
• parallelismParallelization is the execution of independent computational operations concurrently across multiple processing resources.
• column blockA column block is a partition of dataset columns or features that enables parallel processing during model training.
• quadratic approximationA second-order approximation estimates a function locally using both its first and second derivatives.
• second derivativeA second derivative measures how the rate of change of a function itself changes with respect to its input.
• imputationImputation is the replacement of missing data with estimated values or classes derived from available information.
• DecisionTreeRegressorA scikit-learn estimator that fits a decision tree for predicting numeric target values.
• XGBClassifierXGBClassifier is an XGBoost estimator for supervised classification tasks.
• irrelevant featureAn irrelevant feature is an input attribute with little or no association with the target and little useful information for the prediction task.
• service account keyA service account key is a credential containing information, including a private key, for authenticating software as a Google Cloud service account.
• gamma objectiveA gamma objective is a regression loss objective based on the gamma response distribution for strictly positive, typically right-skewed targets.
• containerA software construct that packages an application with its dependencies so it can run predictably across different environments.

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.