Machine Learning Masterclass with Python, TensorFlow, GCP

Assess cyber security threats in AI-driven ecosystems and how fake data sabotages learning, while exploring defenses like data masking and secure identity verification against BEC and malware.

Env setup for machine learning and deep learning: install Python (2.7 and 3.x), Jupyter notebook, NumPy, Pandas, Matplotlib, scikit-learn, CUDA, and Keras for GPU-accelerated workflows.

Learn how skewness and kurtosis describe distribution shapes, from normal to skewed histograms. Use box whisker plot, scatterplots, and correlation coefficients to analyze data relationships.

Master missing data imputation from simple mean or median imputation to regression-based methods, and learn multiple imputation with em algorithm to obtain approximately unbiased estimates under missing at random.

Explore supervised and unsupervised learning, highlighting labeled data, training datasets, and key algorithms for classification, regression, clustering, and association rules.

Explore reinforcement learning by examining an agent in an environment that learns from trial and error through rewards and punishments, guided by states, actions, and maximum reward.

Learn linear regression for mpg prediction using weight and displacement as predictors; fit a regression plane with least squares via normal equations to minimize residuals.

Explore fitting a linear regression model to data using least squares, formulating normal equations, and solving for coefficients with X^T X, while addressing rank and unique solution conditions.

Explore how linear regression uses feature engineering and polynomial terms to model horsepower and mpg, and navigate the bias-variance trade-off with increasing model complexity.

Explore variable selection in linear regression, comparing best subset and stepwise approaches to identify the most predictive features. Use mean squared error and information criteria to select a parsimonious model.

Explore statistical inference in linear regression by examining the sampling distribution of regression coefficients. Test beta with the t statistic and p-values under the null beta equals zero.

Explore logistic regression to model the probability of binary outcomes using covariates as predictors, and apply it to a loan applicant to estimate risk.

Estimate logistic regression parameters using maximum likelihood from the likelihood function for logit and probit variants.

Explore logistic regression accuracy using the confusion matrix—true positives, false positives, true negatives, false negatives—along with maximum likelihood estimation and regularized logistic regression for robust out-of-sample prediction.

Explore decision trees for supervised learning, using a tabular dataset with attributes such as age, job, and house to split data, minimize misclassification, and use purity as a guide.

See how a decision tree uses entropy or the Gini index to measure impurity, reduce misclassification, and build splits by purity gain to classify by majority label.

Explore how impurity gain ratio improves decision tree learning by penalizing high-cardinality attributes like unique identifiers, using entropy and Gini index to measure impurity and guide splits.

Transform numerical attributes into thresholds using queries like length is less than or equal to a constant, illustrated by iris petal length and width; the Gini index guides splits.

Build a decision tree classifier to predict Titanic survival, loading data, preprocessing categorical features, and exploring the survival distribution to estimate a base model.

Build and evaluate a decision tree model to predict outputs with test data, and interpret metrics such as accuracy, recall, specificity, and precision alongside feature importance insights.

Identify feature importance in a decision tree and perform feature engineering on Titanic data by deriving family size, extracting titles with regex, and creating a gender/child feature.

Explore decision tree modeling with grid search cross-validation to tune hyperparameters, improve accuracy, and reveal top features such as gender, child, and family size through feature importance.

regression trees predict numerical values by partitioning inputs into intervals and using the mean within each partition, guided by mean squared as the cost.

Explore partition-based and hierarchical clustering, including k-means, and learn distance metrics like Euclidean, Minkowski, cosine, and correlation. Assess cluster quality and estimate the number of clusters.

K-means clustering partitions data into k clusters by minimizing within-cluster variance. It assigns points to the nearest centroid using Euclidean distance, then updates centers to the mean until convergence.

Explore k-means clustering with Python on the Pima diabetes dataset, learn initialization, convergence, and how to choose the number of clusters using Davies-Bouldin and silhouette indices.

Explore hierarchical clustering: compare agglomerative and divisive approaches, visualize with a dendrobium tree, and learn linkage methods using similarity or dissimilarity metrics to merge clusters.

Apply hierarchical clustering to health care and social networks, tracing viral outbreaks with HIV phylogenetic analysis and analyzing political alliances on Twitter using the Waltrip algorithm and agglomerative clustering.

Learn how hierarchical clustering builds cluster hierarchies by iteratively merging closest clusters using a distance matrix and the single linkage method, from five observations to a final cluster.

Explore hierarchical clustering and its linkage methods—single, complete, average, centroid, and Ward—explaining distance calculations, weighted sums, and their role in initializing k-means seeds.

Learn python-based image classification with TensorFlow on Google Colab, training on 60,000 fashion images and testing on 10,000 images across clothing categories.

Train a neural network on a 60,000-image 28×28 dataset with TensorFlow, preprocess by scaling pixels to 0–1, and compare training and test accuracy to illustrate overfitting and inference in Colab.

Explore how Google's ml and ai innovations shape cloud infrastructure, emphasizing security, encryption, and high-speed networking, plus McQuary enables no-code ml using sequel.

Explore Google Cloud basics via BigQuery, create a dataset within a project, query public data, and build and evaluate a logistic regression model directly in SQL on the cloud.

Connect to a Linux virtual machine on Google Cloud, set up a Python virtual environment, install packages, copy a dataset from a bucket, and train a deep neural network.