AWS SageMaker Practical for Beginners | Build 6 Projects

Learn how to create a free AWS account, understand the AWS free tier limits for SageMaker, EC2, and S3, and explore the console to build, train, and deploy ML models.

Explore the basics of artificial intelligence, machine learning, and deep learning, including data, models, and compute power, with AWS SageMaker and S3 storage for training and deployment.

Explore the fundamentals of AI, machine learning, and deep learning, including supervised, unsupervised, and reinforcement learning. See practical examples like classification, regression, clustering, and reward-based learning, with upcoming data lingo.

Learn ai and ml data lingo, distinguish labeled from unlabeled data, and compare good data versus bad data to train reliable models.

Explore the basics of Amazon Web Services and cloud computing, including pay-as-you-go compute, storage, and machine learning with SageMaker. Learn how AWS unlocks agility, elasticity, and global deployment.

Discover the three core machine learning components—data, model, and compute—using AWS SageMaker, S3 storage, and EC2 compute in a guided management console tour.

Understand AWS regions and availability zones as geographically separated data centers that improve data compliance and reduce latency, and learn how to select regions and zones in the console.

Amazon S3 provides scalable, durable cloud storage for SageMaker data, with buckets, access controls, and lifecycle tiers to store and manage training and testing data.

Explore EC2's resizable cloud compute with standard, memory, compute, and GPU instances for AI training. Learn pricing (on-demand, spot, reserved), elastic inference, and IAM basics for SageMaker security.

Get a high-level look at Amazon SageMaker, a fully managed machine learning workflow that handles data labeling, model training and tuning, deployment to scalable endpoints, and production monitoring.

Explore a high level tour of AWS SageMaker, from labeling data with ground truth and Mechanical Turks to training, tuning, and deploying models via notebooks, endpoints, and the AWS marketplace.

Explore how AWS SageMaker Studio unifies building, training, tuning, and deploying models in a one-stop shop browser IDE, with Autopilot generating and evaluating multiple models from S3 data.

Explore Amazon SageMaker Studio's intuitive interface, create and run notebook-based experiments, upload data to S3, and train multiple models with Autopilot before deploying endpoints for production.

Master AWS SageMaker model training and deployment using S3 data and ECR training images. Create endpoints for inferences and leverage batch transforms for bulk predictions.

Explore the basics of regression with SageMaker by building, training, deploying, and using a linear regression model to predict employee salaries from years of experience.

Learn the basics of simple linear regression, predicting a salary (Y) from experience (X) by fitting a line with slope M and intercept B, for future predictions in AWS SageMaker.

Apply the least squares technique to obtain the best fit line by minimizing squared residuals. Split data into training and testing to ensure generalization.

Learn the basics of the AWS SageMaker linear learner for supervised regression and classification. Includes data shuffling, feature scaling, hyperparameters, and parallel model training.

Learn to load and run Jupyter notebooks in AWS SageMaker by instantiating a notebook or using SageMaker Studio, train models, deploy endpoints, and generate predictions.

Explore Amazon SageMaker Studio, the first fully integrated ml ide, to code, upload notebooks, track experiments, visualize data, and train a linear regression model with scikit-learn.

Import pandas, numpy, and matplotlib in SageMaker Studio, load salary.csv into salary_df, view the first and last seven rows with head and tail, and find the maximum salary.

Perform exploratory data analysis and visualization on salary_df with seaborn, including heat map, describe, histogram, pair plot, and a reg plot linking years of experience to salary.

Create training and testing datasets by splitting input X (years of experience) and output Y (salary) with train_test_split, convert to float32 numpy arrays, and verify a 75/25 split for generalization.

Train a linear regression model in scikit-learn using train/test splits and evaluate with model score. Inspect coefficients and intercept; compare fit intercept true versus false to see the effect.

Evaluate the performance of a trained linear regression model in AWS SageMaker by predicting salaries from years of experience, and visualize the fit with training data and a regression line.

train a linear learner model in AWS SageMaker using S3 for training data and model artifacts. tune hyperparameters and use spot training to improve performance and cut costs.

Deploy and test a trained linear learner model on Amazon SageMaker, invoke the endpoint with text/csv data, review json predictions, and learn endpoint management to control costs.

Predict health insurance costs using multiple linear regression in AWS SageMaker, leveraging features like age, gender, BMI, number of children, smoking status, and region.

Explore intuition behind multiple linear regression, extending simple regression to multiple features X1..Xn with coefficients B1..Bn and a bias, and learn to optimize parameters and evaluate with MSE, MAE, R-squared.

Explore regression metrics and KPIs, including RMSE, MSE, MAE, and MAPE, to assess model fit and error in regression predictions.

Calculate R-squared, the coefficient of determination, to quantify how much of the variance in insurance cost is explained by age; use adjusted R-squared to penalize useless predictors in code.

Build, train, and deploy a regression model to predict insurance charges using SageMaker Studio; import libraries, explore data, and evaluate linear and neural network models.

Perform exploratory data analysis on an insurance dataset. Check for nulls with a heatmap, summarize with info and describe, and compare region charges using grouped means and dummy variables.

Visualize the insurance dataset with histograms, regplot, pair plots, and a correlation heat map to reveal relationships among age, BMI, charges, and other features.

Create training and testing datasets for a regression model by selecting inputs and the charges as output, scaling with standard scaler, and applying a train-test split to evaluate performance.

Practice training and evaluating a linear regression model locally with scikit-learn, including 80/20 train-test split, data scaling, fitting, predicting, inverse transforming, and metrics such as RMSE, MSE, MAE, and R-squared.

Train a linear learner model in AWS SageMaker by converting data to record IO, uploading to S3, and configuring a built-in estimator with key hyperparameters to run the training job.

deploy a trained linear learner model in AWS SageMaker, create and test an endpoint for inference with text csv input, and review predictions and metrics while noting potential hyperparameter tuning.

Explore artificial neural networks, their basic neuron model with inputs, weights, bias, and activation functions, and train them for regression tasks.

Analyze activation functions sigmoid, relu, and tanh, showing sigmoid for output probabilities, relu for non-linearity in hidden layers and regression, and tanh for zero-centered classification outputs.

Explore multilayer perceptron networks by building two-neuron and multi-layer dense nets. Learn matrix multiplication of inputs and weights, biases, activation functions, and training to minimize error.

Explore how artificial neural networks train through supervised learning, updating random weights with the error signal, while training, testing, and validation data manage over-fitting and generalization.

Apply gradient descent to minimize the loss function and update neural network weights and biases using a learning rate, guided by gradients and step size, toward optimal minima.

Learn how backpropagation trains neural networks by performing forward propagation, computing error, propagating it backward, and updating weights with a learning rate to adjust gradients and deltas.

Train a dense neural network for regression with the Caris API on TensorFlow, using dropout and ReLU, and the Adam optimizer. Evaluate with mean squared error and report RMSE, R-squared.

Learn to forecast weekly retail store sales with XGBoost regression in SageMaker, using a three-file dataset of stores, features, and sales, and tune hyperparameters for accuracy.

Understand bias-variance trade-off by comparing simple and complex models, training vs testing data, and how to seek an optimal balance to improve generalization and avoid overfitting.

Learn how regularization prevents overfitting by penalizing model complexity, focusing on L2 ridge regression, alpha controls bias-variance, and how a flatter slope improves generalization across training and testing data.

Explore l1 regularization (lasso) in regression, adding alpha times absolute slope to reduce overfitting and enable feature selection, and compare it with l2 regularization for when to drop features.

Explore XGBoost, an extreme gradient boosting algorithm using an ensemble of weak models to improve regression and classification, with SageMaker hyperparameter tuning and distributed training.

Explore boosting, an ensemble machine learning approach that learns from previous mistakes by training sequential weak models, reweighting data based on residuals, and combining predictions for stronger outcomes.

Explore decision trees and ensemble learning, including boosting and XGBoost, and learn how CART trees split data at decision nodes and leaves to combine predictions for robust outcomes.

Explore gradient boosting and XGBoost, learn how residuals drive ensemble trees, apply learning rate and regularization to improve generalization and reduce overfitting.

Discover XGBoost, a scalable gradient boosting system with sparse data handling and weighted quantile sketch, enabling fast, end-to-end training on billions of examples with SageMaker.

Explore SageMaker XGBoost, an ensemble algorithm robust for regression and classification, and learn data preparation, csv or libsvm formats, and core hyperparameters like max depth, gamma, eta, alpha, and lambda.

Apply XGBoost to regression tasks using Amazon SageMaker notebook instances, not SageMaker Studio, tune hyperparameters, train and deploy models, and forecast weekly sales from multiple datasets.

Load and explore stores, features, and sales datasets with Pandas, NumPy, and visualization tools to forecast weekly sales as a regression target, review holidays and markdowns, and prepare datetime features.

Merge sales, feature, and stores data frames with pandas on store, date, and is_holiday; then extract the month from date into a new months column.

Explore the merged dataset, check for nulls with a heat map, fill nulls with zeros, describe statistics, and convert boolean holidays to binary.

Perform exploratory data analysis with pivot tables to compare weekly sales across store types A, B, and C, and visualize markdowns and holiday effects using heatmaps and histograms.

Prepare regression data by splitting inputs from weekly sales, dropping the date, one-hot encoding categoricals with drop-first, and converting to float32 arrays for an 85/15 train-test split for XGBoost.

Train a local XGBoost regressor with squared error, evaluate with r-squared and rmse, and tune learning rate, max depth, and estimators for model performance.

Train an XGBoost model using SageMaker. Prepare train and validation data as csv, upload to S3, configure the container and hyperparameters, and run fit to produce artifacts.

Deploy an XGBoost regressor as an endpoint, feed data in text and CSV format, run batch predictions in 10,000-sample chunks, and evaluate with RMSE, MAE, and R-squared before deleting endpoint.

Learn to run a hyperparameters tuning job in SageMaker, compare random and Bayesian searches, and optimize XGBoost hyperparameters like learning rate, max depth, and rounds using training and validation data.

Train the model using optimized hyperparameters from tuning, redeploy the SageMaker endpoint, and evaluate predictions with RMSE, MSE, and R-squared on partitioned test data.

Leverage principal component analysis for dimensionality reduction and apply XGBoost to classify cardiovascular disease using AWS SageMaker. Analyze age, blood pressure, cholesterol, and glucose to assess risk.

Explore the intuition of principal component analysis for dimensionality reduction using SageMaker PCA, its regular and randomized modes, pipe mode, and then train an XGBoost classifier on the reduced components.

Explore XGBoost for classification tasks, an ensemble gradient-boosted tree method that combines multiple models for robust predictions, with applications like fraud detection and retirement eligibility.

Explore how the confusion matrix visualizes binary classification performance, defining true positives, true negatives, false positives, and false negatives, plus accuracy, precision, recall, and type I and II errors.

Explore precision, recall, and f1-score, and learn why accuracy can mislead on unbalanced data. Use real-world examples—from fraud detection to spam filtering—to understand when to prioritize each metric.

Understand roc curves and auc as metrics for performance. Analyze true positive rate versus false positive rate across thresholds, and compare the area under the curve to a random predictor.

Explain the difference between overfitting and underfitting, show how simple linear models underfit nonlinear data, and complex models memorize training data but fail to generalize, guiding toward balanced generalization.

Set up a SageMaker Studio notebook, perform exploratory data analysis, and apply PCA for dimensionality reduction before training and deploying an XGBoost classifier with endpoint inference.

Learn to import Pandas, NumPy, Seaborn, and Matplotlib, load dataset, drop id column, convert age from days to years, and perform exploratory data analysis for a cardiovascular disease classifier.

Visualize the dataset with histograms and a correlation matrix, then prepare training and testing data by separating features from the cardio target for modeling.

Train and test an XGBoost classifier in local mode, explore default settings, and perform grid search hyperparameter optimization across gamma, subsample, colsample_bytree, and max_depth.

Perform dimensionality reduction with the built-in PCA in AWS SageMaker, apply unsupervised learning, prepare data, upload to S3, and train a PCA model with components.

Deploy a trained PCA model as an endpoint, feed input data to obtain six principal components from eleven features, and retrieve JSON-formatted predictions to prepare for the next task.

Train a built-in XGBoost classifier in SageMaker for binary classification using training and validation splits, plus PCA dimensionality reduction features and CSV data pipelines.

Deploy the trained XGBoost endpoint, run inference on test data, extract predictions, compare to ground truth, and visualize precision, recall, and accuracy (about 72%), plus the confusion matrix.

Learn to build and deploy a multi-class image classifier using convolutional neural networks in SageMaker, applying Fashion-MNIST and German Traffic Sign data to recognize 10 and 43 classes.

Learn the basics of convolutional neural networks and how they learn by applying convolution, relu non-linearity, pooling, and flattening into dense networks for image classification.

Learn how convolutional neural networks extract features with convolution, relu, pooling, and flattening, then classify via dense layers for image recognition.

Explore two strategies to improve CNN performance: adjust the number of filters and apply dropout to improve generalization across training and testing data.

Explore the confusion matrix as a visual tool to evaluate a multi-class traffic sign classifier, detailing true/false positives and negatives, and calculating accuracy, precision, and recall.

Explore the LeNet convolutional neural network used for image classification, detailing its C1 convolution, S pooling layers, a second convolution with 16 filters, and dense layers for 43 outputs.

Learn how to request a service limit increase for SageMaker notebooks to access GPU instances like P216 X Large, including step-by-step AWS console instructions and a 24–48 hour approval timeline.

Build a deep learning traffic sign classifier in SageMaker Studio by uploading data, preparing train, validation, and test sets, and visualizing sample images to validate data quality.

Import SageMaker and Boto3, create a session and execution role, prepare training and validation data as npz files, and upload them to S3 under the traffic sign classifier prefix.

Train a Lunette CNN on SageMaker to build and train a 43-class classifier with 32×32×3 inputs, five-by-five filters, pooling, and dense layers.

Deploy a trained SageMaker model as an endpoint and generate predictions on unseen traffic sign images. Evaluate predictions against true labels, and delete the endpoint to avoid charges.

Explore how ai transforms the insurance industry—from risk assessment, marketing, sales, and service operations—and learn to predict health insurance costs using aws SageMaker Studio Auto Pilot and regression models.

discover ai in business with AWS SageMaker Autopilot, which automates feature engineering, data cleaning, model building, training, tuning, and deployment to predict insurance premiums as a regression task.

Explore predicting individual health insurance charges using age, gender, bmi, children, smoking, and region; feed data to an AI model and explore risk modeling and regression analysis in insurance.

Explore simple and multiple linear regression to predict insurance costs from factors like age and other inputs, learning the equations, slope and intercept, and least-squares estimation.

Discover AWS as a leading cloud platform, learn essentials of storage, compute, and machine learning with SageMaker Autopilot, and explore building AI models in the cloud without upfront hardware costs.

Discover how AWS S3 provides scalable storage with buckets and 11 nines durability, and how EC2 delivers resizable compute power for training models with SageMaker.

Explore AWS SageMaker, a managed machine learning workflow for labeling, training, tuning, and deploying models with minimal coding. Use SageMaker Studio with autopilot for hyperparameter optimization and near zero coding.

Evaluate regression models by analyzing residuals and using MAE, MSE, RMSE, and R-squared to measure prediction accuracy for a linear regression predicting insurance cost from age.

Explore creating an AWS SageMaker Autopilot experiment, upload your insurance dataset to S3, specify the target and output locations, and review multiple model runs to find the best model.

Create an AWS SageMaker Studio AutoPilot experiment by specifying your data in S3 and the target column charges, then let AutoPilot perform preprocessing, feature engineering, and model tuning for regression.

Run an autopilot workflow in AWS SageMaker Studio, review the candidate generation and data exploration notebooks, evaluate models by mean squared error, and deploy scalable endpoints.