Machine Learning on AWS SageMaker for Beginners

Explore how machines learn from input features and labeled data, train models with algorithms, and use test data to make predictions.

Compare traditional programming with machine learning by detailing algorithm development, coding, and results versus data collection, model training, and prediction, highlighting outputs of processed data versus a trained model.

Explore the six-step, iterative machine learning workflow—data collection, cleaning, visualization, model selection, evaluation, and deployment—and learn how data quality and data visualization relate to data imbalance.

See how machine learning powers daily apps like Facebook ads and search, Amazon recommendations and price variation, Netflix suggestions, gaming innovations, and UBA route and fare predictions.

Explore supervised learning algorithms, including linear and logistic regression, decision trees, k-nearest neighbors, and naive Bayes, with emphasis on regression targets, classification, and model fitting.

Explore unsupervised learning with k-means clustering and fuzzy clustering concepts, where cluster centers are means and points have degrees of membership that drive iterative improvements.

Explore reinforcement learning with q-learning, q values, states and actions, and the temporal difference update rule. Examine episodes, greedy policy with epsilon, gamma, and alpha, and model-free trial-and-error learning.

Create and verify an AWS account with billing and contact details, authenticate via a phone verification code, and access the AWS management console to explore analytics and Amazon Redshift.

Explore the AWS web console to access services, launch virtual machines or build web apps, and manage billing, security, and region settings across compute, storage, databases, and more.

Learn to create a SageMaker notebook instance, choose instance type and settings, configure permissions, encryption, and VPC options, and manage uptime to avoid charges.

Spin up a Jupyter notebook in AWS SageMaker by starting the notebook instance, stopping when idle to save costs, opening JupyterLab, and running basic Python commands.

Upload your Titanic dataset to a new S3 bucket, configure region and public access, create a dataset folder, and upload files for use in SageMaker projects.

Import data from S3 into a Jupyter notebook on AWS SageMaker using pandas, then load a CSV dataset like Titanic for exploratory data analysis and model training.

Develop a basic Titanic survival prediction project by transforming a passenger dataset, handling missing values, and performing feature engineering to train a model in a Jupyter notebook.

Import the Titanic dataset from S3 into a SageMaker notebook, load it as a dataframe, and prepare for exploratory data analysis.

Explore exploratory data analysis on Titanic dataset by inspecting data frame info, describing statistics, and counting survivors and gender to prepare for cleaning, feature engineering, and model training on SageMaker.

Perform data cleaning on a Titanic dataset by extracting titles, converting gender to numeric, imputing missing ages, and categorizing data, followed by preprocessing analyses and visualization to guide model training.

Convert gender to numeric values 1 and 0, fill missing ages, categorize passengers by age, and drop unused columns to prepare a clean 891-row by 13-column dataset for model training.

Split the dataset into training and test sets, define features X and output Y, convert data to float32, and use a 30% test split to evaluate the model.

Train a machine learning model in AWS SageMaker using the linear learner supervised learning algorithm, with a notebook, execution role, and training and validation data.

Deploy the trained model to a SageMaker endpoint by calling the deploy method, specify the instance count and type, and learn about docker images and serializers used for inference.

Learn to run survival predictions with a SageMaker model, pass test data, interpret scores and predicted class, and delete endpoints to avoid charges — geared for beginners.

Define the Boston housing price problem and import the dataset with AWS SageMaker for beginners. Explore, preprocess, train, deploy, predict prices, and finally delete the endpoint.

Conduct exploratory data analysis on a housing dataset to identify key feature correlations with price, select four influential features, and visualize relationships to guide final model training.

Master univariate and multivariate analysis, comparing single-variable description with multi-variable hypothesis testing. Examine price data using box plots and distribution plots, and apply the interquartile range method to handle outliers.

The lecture continues univariate and multivariate analysis, detects outliers, assesses normality and correlations, removes suspect observations, and prepares a cleaned dataset split into training and test sets for modeling.

Split the data into train and test sets, prepare X and Y, convert to float32, and train and validate a model with 30 percent test size and a random state.

Train a linear regression model on AWS SageMaker using the linear learner library for price prediction, configuring training jobs, data channels, and deployment endpoints.

Predict house prices with a deployed sageMaker endpoint using a json serializer for input and output, compare results to actual values, and delete the endpoint to avoid charges.

Apply principal component analysis to US county demographics to perform population segmentation and uncover natural county clusters. Import and explore the data from S3, preparing for unsupervised learning.

Perform exploratory data analysis on the counties dataset, drop incomplete rows, summarize statistics, visualize histograms for key occupations, and prepare data with min-max scaling for PCA.

Apply principal component analysis to reduce data dimensionality and build a SageMaker PCA training job by preparing float32 features, configuring resources, and fitting the model.

Learn to access PCA model attributes, extract principal components and singular values, compute explained variance for the top components, and interpret each component's makeup.

Deploy the principal component analysis model on SageMaker and configure endpoints to transform county data into five components. Interpret the components to reveal patterns in poverty, unemployment, self-employment, and income.

Apply the k means algorithm to segment the population of counties using five transformed components, prepare data for SageMaker, and train a five-cluster model on the transformed counties dataset.

Explore a k-means model's attributes by examining cluster distributions, centroids in the transformed space, and heat maps of PCA components. Map counties to clusters and interpret cluster characteristics.

Conclude the unsupervised learning workflow by clustering a data set with PCA-based dimensionality reduction and SageMaker models, then delete endpoints to avoid AWS charges.

Explore building a handwriting digit recognition model using the amnesty database in SageMaker, starting with environment setup, library imports, and configuring the S3 bucket for models.

Automate the download and import of the dataset into a Jupiter notebook, unzip the zip gz file, and load the training, validation, and test sets using latin-1 encoding.

Explore the handwritten image training dataset, viewing sample images and their labels from the 50,000 training images, to prepare for training with the exposed algorithm.

Transform the training data into CSP format for XGBoost and save train, validation, and test sets to S3, enabling the optimized distributed gradient boosting workflow.

Train the model with the XY boost algorithm by importing libraries, defining the data template, and setting the train, validation, and output paths, then tune parameters and begin training.

Deploy and validate a model on aws sageMaker by loading test data from s3 and predicting handwritten digit images with XY boost algorithm. Clean up deployed endpoints to save costs.