DY0-001: CompTIA DataAIDemo

The minimum conventional architecture of an artificial neural network includes an input layer, one or more hidden layers for learned feature transformations, and an output layer that produces the model’s result. Convolutional, pooling, and dropout layers are optional specialized layers.

Cloud deployment provides on-demand access to scalable computing resources, allowing an organization to increase or decrease capacity as required.

Constrained optimization is used when an objective, such as minimizing transport time or fuel use, must be optimized subject to limits on available resources. The 125-gallon fuel limit is an explicit constraint on the feasible transportation plan.

The hyperbolic tangent (tanh) activation function has an output range of −1 to 1, giving it a negative lower bound.

An isolated environment separates the deployment service from internal servers, meeting the security requirement while allowing the model to be publicly available.

Convolutional neural networks (CNNs) are designed for image data: convolutional layers learn local spatial patterns and build them into higher-level visual features for recognition tasks.

A literature review establishes the relevant prior research, candidate approaches, known constraints, and feasibility considerations needed to scope a machine-learning proof of concept. Model selection, hyperparameter tuning, and performance evaluation follow once an approach has been identified and implemented.

Geocoding transforms a physical address into geographic coordinates, such as latitude and longitude, so the location can be plotted on a spatial map.

There are 165 correct predictions, producing 165 hours saved, and 35 incorrect predictions, producing 140 hours lost. The resulting net impact is 25 hours saved.

The expression anchors the entire value, requires 3 as the first digit and either 4 or 7 as the second digit, then permits exactly 13 more digits. This produces a total length of 15 digits and allows only prefixes 34 and 37.

A Sankey diagram displays how quantities flow between categories, with link widths representing their relative values. It is well suited to showing the components and movement of departmental revenue and cost of goods sold.

CI/CD automates the build, testing, and deployment pipeline, enabling repeatable, iterative software releases.

Positive-versus-negative sentiment assigns each text review to one of two labels, making it a binary classification task. A binary classifier can be trained on labeled reviews to predict either sentiment category.

A neural network uses input features, learned weights, and bias terms to compute outputs for classification tasks.

Data compression requires processor work to encode the dataset, and subsequent processing commonly requires decompression. This added computation can increase server CPU utilization.

Monte Carlo simulation models uncertain, time-evolving outcomes by repeatedly sampling possible transitions and event paths. It is therefore suited to representing how an individual may move through multiple major life events under stochastic variation.

A container bundles the model, runtime, libraries, and dependencies into a consistent, portable unit. Managed container platforms can run that unit without requiring management of the underlying servers or orchestration infrastructure, reducing deployment and operational effort.

KNN uses the nearest labeled data points to assign a class to a new observation, whereas k-means groups unlabeled observations into clusters by iteratively determining cluster centroids.

Model performance targets must be aligned with the business outcome and agreed with business stakeholders. Once the model meets the stated minimum requirement, any additional accuracy work must be evaluated against the delivery deadline and its business value; the key stakeholder should determine whether changing the agreed tradeoff is warranted. AWS guidance states that stakeholders should agree on performance metrics and that technical metrics must be connected to business objectives.

Backpropagation sends the loss error gradients backward through the network, enabling calculation of how each weight and bias contributed to the error and allowing those parameters to be updated.

Regular expressions are designed to identify and extract text that matches defined patterns, including components embedded in URLs or other website-address strings.

Binning partitions continuous age values into defined ranges, such as age groups, so observations can be aggregated at the group level.

Multivariate outliers are records whose combination of values is unusual across several variables, even when no single value is extreme. An individual's income, assets, age, profession, education level, and address may form atypical combinations that can distort a predictive model. Outliers can negatively affect model accuracy and should be detected during data preparation.

A systematic error is a consistent, repeatable bias in measurements, such as a sensor that persistently reads too high relative to comparable sensors.