Mastering Artificial Intelligence (AI) with Python and R

Explore supervised and unsupervised learning, including regression, classification, and clustering, with practical examples and a hands-on life cycle from data preprocessing to model retraining.

Explore bias, variance, and their trade-off; learn how training-test splits affect model performance, and how to avoid overfitting and underfitting.

Load the housing data with pandas read_csv, handle missing values by imputing the mean for total bedrooms, and encode the ocean proximity. Scale the data to standardize for regression readiness.

Check visualizations with heat maps and box plots to assess distributions and outliers. Build and evaluate a baseline linear regression model using mean squared error and R-squared.

Apply principal component analysis and linear discriminant analysis to reduce dimensions of iris data, extracting two components and visualizing PCA and LDA results.

Explore model selection for iris classification by preparing data, dropping the id column, encoding the species with label encoding, and performing a stratified train-test split.

Assess dataset statistics with describe to view quartiles, mean, std, and count; examine status id and status published for uniqueness, then prepare for unsupervised learning via clustering with label encoding.

Use a label encoder to convert the status type to categorical labels and apply the min max scaler, preparing data for a k means clustering model and evaluating inertia.

Assess labeling accuracy by computing the accuracy score on the labeled data, then apply the elbow method to determine the optimal number of clusters for k-means.

Learn to set up and run TensorFlow in Jupyter, build a tf.keras sequential model, compile with loss and metrics, and apply the deep learning life cycle—from define to predict.

Learn to build a binary classification model with the TensorFlow functional API, including data loading, label encoding, train-test split, dense layers, training, evaluation, and prediction.

Explore heat maps, box plots, and pear plots; define x and y data, perform train-test splits, and compare linear regression with random forest using mean squared error and r-squared.

Explore random forest and extremely random forest as ensemble learning methods built from decision trees with random data subsets, and learn their boundaries, classification report, and confidence measures.

Explore how grid search cv finds optimal classifier parameters by testing a parameter grid with cross-validation, comparing precision and recall to select the best settings.

This lecture demonstrates predicting traffic using an extremely random forest regressor with label encoding, evaluating with mean absolute error, and predicting vehicles for a specific date, time, and location.

Explore unsupervised learning to detect patterns in unlabeled data by clustering with k-means, using similarity metrics to form subgroups and iteratively update centroids.

Learn how the meanshift clustering algorithm, a non-parametric method, identifies cluster peaks by shifting centroids within windows until convergence, with applications in object tracking and real-time data analysis, using Python.

Load 2d data from a csv, import numpy, matplotlib, sklearn, and apply meanshift clustering with bandwidth estimated by quantile for kernel density estimation.

Learn affinity propagation, a clustering algorithm that does not require pre-set cluster counts, using message passing to identify exemplars through responsibility and availability. Apply to stock market data with Python.

Use silhouette scores to evaluate clustering quality and identify the optimal number of clusters with a Python-based visualization.

Learn gaussian mixture models, a weighted sum of gaussian components that model multi-modal data. Apply to image database retrieval, stock market fluctuation, biometric verification, with Python examples.

Build and visualize a Gaussian mixture model using the iris data with an 80/20 split, training a GMM classifier and plotting ellipses for three distributions.

Explore the Naive Bayes classifier built on Bayes theorem with an independence assumption, implement Gaussian NB in Python, and evaluate with cross-validation and metrics like accuracy, precision, recall, and F1.

Create and visualize a confusion matrix in Python using NumPy, Matplotlib, and scikit-learn, compare true and predicted labels, and generate a classification report with precision, recall, and F1 scores.

Learn how support vector machines classify data by an optimal hyperplane with maximum margin, using support vectors, and apply Python examples to predict income brackets from 14 attributes.

Parse a family tree with logic programming in Python using Kanren and a JSON data set. Define parent, grandparent, sibling, and uncle relations and query children, grandchildren, and spouses.

Learn heuristic search for solving problems by eliminating obviously wrong options to speed up solutions. Compare uninformed and informed search, with depth-first, breadth-first, and uniform-cost techniques.

Learn to build an eight puzzle solver with the a-star algorithm, using a heuristic search cost function and Manhattan distance to select the most promising path toward the goal.

Solve the region coloring problem with heuristic search in a constraint satisfaction framework (CSP) using Python, coloring regions with four colors so adjacent regions differ.

Explore examining text with nltk using concordance to show word occurrences with context. See how similar shows words in context and common context reveals shared context between words in Python.

Access and tokenize raw text from local files or web sources using Python. Learn to use urllib, urlopen, read, and NLTK's word_tokenize to produce tokens.

Explore how stemming reduces inflected words to base forms using NLTK, and compare Porter and Lancaster stemmers through Python tokenization and stem application.

Explore segmentation in natural language processing, contrasting sentence segmentation and word segmentation, and learn how tokenization uses a segmenter such as a Python-based sentence segmenter, with challenges from abbreviations.

Explore information extraction in natural language processing, converting unstructured text to structured data, and applying steps—sentence segmentation, tokenization, pos tagging, entity and relation recognition—using Python and NLTK.

Describe tag patterns as rules for chunk grammar using angle brackets and part-of-speech tags to define optional, zero-or-more, and one-or-more sequences for parsing sentences.

Learn iob text tagging for chunks with i, o, b labels and chunk types, and compare with tree representations where s marks a sentence and np denotes noun phrases.

Apply chunking with regular expressions using NLTK's regex parser to build chunk structures from grammar rules, observe leftmost precedence for overlapping matches, and generate noun phrase chunks.

Explore named entity recognition, known as entity chunking and extraction, as a subtask of information extraction that locates boundaries and identifies types such as persons, organizations, locations, times, and quantities.

Define artificial intelligence as the science and engineering of making intelligent machines and intelligent agents; the Turing test guides objective evaluation and modern AI emphasizes practical problem solving.

Explore how intelligent agents perceive environments, map the percept sequence to actions, and design rational agents considering environment properties.

Explore intelligent agents that perceive environments through sensors, act via actuators, and use heuristic search, learning, planning, and knowledge representation to solve problems.

Trace the backtracking algorithm on a graph using the given pseudo code; start at a, generate b, c, d, then explore b, e, f, and backtrack on dead ends.

Illustrates backtracking on a graph by managing the state list, the next-state list, and the dead end, while generating children and tracing the path to the goal gca.

Explains state-space search using backtracking, depth-first and breadth-first search, detailing nodes, states, and the role of the state list and NSL in iterative search.

Explore heuristic search for problem solving, including defining heuristics, their role in guiding state-space search, and methods like best first search and pruning to avoid combinatorial explosion.

Apply a heuristic calculation technique to tic tac toe by counting possible wins from x's position, then select the highest heuristic value to guide next moves using f(n)=g(n)+h(n).

Explore best first search, using an open and close list with a heuristic function f(n)=h(n) to greedily reach the goal via a priority queue, maintain ancestor data and manage duplicates.

Apply best first search to the eight puzzle by maintaining open and closed sets, expanding the lowest f(n)=g(n)+h(n) state and generating children until the goal state is reached.

apply best-first search with f(n)=g(n)+h(n) to the 8-puzzle, computing g and h (tiles out of place) and selecting lowest f for expansion; discuss admissibility, optimality, and performance.

Learn how alpha beta pruning enhances the min max search by using a depth-first approach with alpha and beta values to prune branches in two-person games, reducing the search space.

Learn the perceptron learning algorithm for a single-layer network with inputs, weights, and a threshold; update weights via the learning rate and error (t-y), noting non-linear separability limits.

Explain how perceptrons linearly separate data and why xor is not linearly separable, requiring multi-layer networks; introduce sigmoid activation and lambda's effect.

Explore backpropagation in multilayer networks, propagating output errors backward to adjust weights via the delta rule, gradient descent, mean squared error, and the logistic activation function.

Compute hidden node delta w and propagate errors through layers using backpropagation, applying sigmoid activations and iterative weight and bias updates with a learning rate.

Apply backpropagation by initializing weights, forward-propagating inputs, computing and propagating errors for output and hidden units, and updating weights and biases with a learning rate until a threshold is met.

Update weights and theta using learning rate and error signals to adjust neural network parameters, then explore clustering with K-means and unsupervised learning, minimizing intra-cluster distance and maximizing inter-cluster distance.

Explore logic and reasoning, including reasoning, logical inference, and logic languages, with propositional calculus and predicate calculus, and see how the knowledge base and inference engine derive conclusions.

Explore propositional calculus concepts, including atomic symbols p, q, r, and truth values true and false, and learn how negation, conjunction, disjunction, implication, and equivalence form sentences.

Explore the predicate calculus, a more flexible logic with predicate variables and quantifiers, and its symbols, constants, variables, functions, and arity, and how sentences are formed.

Convert English statements to predicate and clause forms, apply negation, use resolution refutation, and show variable cancellation to determine if John is happy.

Convert statements to predicate and clause formats, then perform a resolution proof. Use unification and substitution to derive the null clause and conclude an exciting life (John is exciting).

Explore the knight's tour on a 3x3 chessboard by mapping moves to production rules of conditions and actions, then trace an iteration from 1 to 2 using highest-priority choices.

Explore rule matching and conflict resolution in production rules, including refraction, recency, and specificity, with data-driven and goal-driven chaining in clips and working memory.

Learn to build rule-based systems with clips by defining facts, watching and resetting them, and creating diff rules that map patterns to actions, with testing and loading capabilities.

Open CLIPS and learn to manage a fact database by asserting facts such as color green, listing them with the facts command, and retracting or clearing as needed.

Explore asserting facts in CLIPS, noting case and space sensitivity, and learn to manage facts with retract, watch facts, and memory indicators.

Learn to use clips to run and assert facts like animal is duck and sound is quack, using forward chaining. Load .clp files to reuse rules.

Examine using defects to preload facts and the reset mechanism to reintroduce them. Apply wildcards and variables in rules to match animal facts and print 'animal found' for each match.

Explore how a rule-based AI uses multiple patterns, facts, and assertion rules to infer mammals, using warm-blooded and egg-laying patterns with diff rule.

Learn to retract facts that start with mammal in a CLIPS-style rule system using diff rules, and apply logic and maths operators to convert into CLIPS format and evaluate expressions.

Learn advanced wildcard pattern matching in CLIPS, using a single question mark and multi-field dollar to extract band names and members from facts, then print results via a DF rule.

Learn how to perform wildcard matching with multi-field wildcard entries, using dollar and dollar question mark wildcards to print band members and their bands from a fact database.

Learn to use the def template to populate slots with personal data, apply the assert statement, and link multiple people with def templates, handling multi-slot values like blood pressure ranges.

Learn to build a deftemplate with multiple slots for personal data, assert facts with deffacts, manage date of birth fields, and apply logical conditional elements for truth maintenance.

Design and debug diff rules and def templates within a personal data and cardiac risk scenario, using trial-and-error learning to assert facts into the fact database.

Learn how a model-based simple reflex agent uses internal state to handle partial observability, updating state from percepts and rules to choose actions.

Explore goal based agents that use explicit goals and reasoning to select actions, contrasting with model based reflex agents, and highlighting search and planning to achieve goal-driven behavior.

Compute maximum expected utility by evaluating actions across states with probabilities and utilities, and explore utility theory, preference axioms, and the value of information.

Explore Markov decision processes, dynamic decision networks, and partially observable MDPs to understand policies, value iteration, policy iteration, and linear programming value iteration, game theory, and mechanism design under uncertainty.

Explore Bayesian rule for conditional probability through examples of sequential draws without replacement, joint and conditional probabilities, random variables, and joint distributions.

Explore how data from environments and sensors drives machine learning, turning raw data into intelligent action through data input, abstraction, and generalization, while considering ethics and privacy.

Collect, preprocess, and explore data to support effective machine learning outcomes; apply missing value imputation, outlier detection, and exploratory data analysis, then train and evaluate models.

Explore supervised learning by contrasting regression and classification. Learn regression types such as linear and non-linear, and classification algorithms like decision trees and k nearest neighbor, plus unsupervised clustering.

Navigate RStudio's interface, manage history and environment, and use preloaded data sets, packages, and help to perform data manipulation in R.

Master simple linear regression with ordinary least square method to estimate the intercept and slope, and interpret r squared, t statistics, p values, and residuals for predictive modeling.

Explore how t and f statistics assess a simple linear regression, including p value, residuals, r square, and interpreting model coefficients like 2.2 and 0.6.

Learn a simple linear regression in R using x and y, fit an ordinary least squares model with lm, and read the coefficients and model summary.

Explore how to compute r square and standard error in a simple regression with x and y, using mean, estimated y, and adjusted r square to assess model fit.

Learn variance, covariance, and correlation with their formulas, explore how they describe relationships between variables, assess linear regression, multicollinearity, and hypothesis testing with p values.

Analyze covariance and correlation between X and Y, build a simple linear regression model, and test correlations with p-values; examine discrete and continuous distributions and related R and Python functions.

Explore normal distribution concepts, including mean and standard deviation, and compute binomial probabilities with R functions dbinom and pbinom, illustrated by practical widget examples.

Explore the Poisson distribution, its lambda parameter, and how events in a fixed time follow a Poisson model, with connections to normal approximations, quantiles, and practical computation in R.

Explore standard normal distribution concepts, central limit theorem, and six sigma using z-scores, p-values, and R's p.test to infer drug effects.

Learn to perform multiple linear regression in R with the lm function, visualize data, and select regression variables using the cars 93 dataset and associated preparation steps.

Explore a multiple linear regression model, interpreting coefficients and p-values for city mpg, train with lm, and validate with R-squared while splitting data 70/30 and comparing models with AIC.

Learn stepwise linear regression with backward and forward selection using AIC, as the lm function iteratively adds or removes variables to reach the best model.

Explore regression variable selection using forward, backward, and both directions, build linear models with lm, compare models via AIC and RSS, and predict mileage per liter with real data.

Explore generalized least squares with air quality data to model ozone using wind and date, assess correlation and autocorrelation, and compare gls variants with predictions.

Learn kNN classification by labeling a new item using majority vote of its k nearest neighbors and distance measures such as Euclidean, Manhattan, Minkowski, Hamming, cosine, and Jaccard.

Learn to measure distances with Hamming distance, Manhattan distance, Euclidean distance, and cosine distance in R, then apply KNN steps on Iris, including normalization, train-test split, and evaluate misclassification rate.

Train a decision tree on car seat data, converting sales indicator to a factor and predicting sales category. Use k-fold cross-validation to prune and avoid overfitting, selecting size 21.

Prune the decision tree to reduce misclassifications and leaves, compare prune models against test data, and discuss post-pruning benefits for avoiding overfitting.

Explore random forest, an ensemble of many decision trees built by bootstrap sampling of data and random feature selection to improve accuracy, using voting for classification or averaging for regression.

This lecture covers preparing a ctg csv dataset, converting nsp to a factor, and building a random forest with training and test splits to evaluate accuracy, sensitivity, and specificity, overfitting.

Learn to evaluate random forest performance by extracting variable importance from fitted models, interpreting Gini-based rankings, and understanding how removing features like class or ltv affects accuracy.

Learn how k-means clustering, an unsupervised algorithm, partitions data into k homogeneous groups using nearest centroids, updates centroids, and employs distance measures and the elbow method to select k.

Apply k-means clustering and the elbow method to select an optimal k, balancing within-group homogeneity and heterogeneity in a 30,000-record social network dataset, with suggested k around 3 or 4.

Apply k-means clustering to a 30,000-record data set to assign cluster labels, uncover dominant interests and demographics, and translate insights into targeted advertising strategies.

Apply aggregation functions to reveal cluster demographics, including mean age, gender patterns, and friends counts, using k-means with within, between, and total sum of squares and R square.

Learn how Naive Bayes classifier uses Bayes theorem to compute posterior probability, assuming class conditional independence, and apply it to spam filtering and text classification.

Learn how to build Bayes theorem components from a frequency table, derive a likelihood table, and apply Naive Bayes to text classification, using spam vs ham and words like Viagra.

Explain how Naive Bayes classifies messages by multiplying conditional feature probabilities with the class prior, and how Laplace smoothing prevents zero likelihood from ruining the posterior.

Apply text mining in R to prepare SMS messages (spam/ham) for classification by reading CSV, cleaning text, building a corpus, and generating a document term matrix.

Create and analyze a term document matrix from SMS data, split into training and test sets, and visualize spam vs ham words with separate word clouds in R.

Create a sms dictionary from training data, filter the document term matrix to frequent terms, convert counts to yes/no, and apply naive Bayes with E1071 for spam detection.

Explore Naive Bayes classification for text in R, building an sms spam filter from training data, evaluating with cross tables, and improving accuracy with Laplace smoothing.

Explore support vector machines as a black box method that uses a maximum margin hyperplane to separate data for binary classification, with vectors, high-dimensional spaces, and practical applications.

Explore how the Gaussian RBF kernel drives SVM-based OCR for classifying 26 letters using 16 features. Train on 16k and test on 4k, noting trial-and-error kernel choices.

Explore how Gaussian RBF kernels improve OCR with SVM on a 16-feature letter dataset, comparing linear and RBF kernels, achieving 84% to 93% accuracy and tuning via E1071.

Explore support vector machines, tuning cost and gamma to maximize accuracy; select the best model from multiple SVM variants and predict labels, achieving around 97% accuracy on test data.

Study dimension reduction techniques by comparing feature selection and feature extraction. Explore methods such as wrapper, filter, embedded, and MRMR, plus PCA applications like data compression.

Transform data with PCA as a statistical procedure based on orthogonal transformation that yields linearly uncorrelated principal components for dimension reduction. Cover eigenvalues and eigenvectors, variance and covariance in PCA.

Learn to perform PCA by mean-centering data, computing the covariance matrix, extracting eigenvalues and eigenvectors, and projecting data onto principal components for dimensionality reduction while preserving variance.

Compute eigenvalues and eigenvectors of a 2x2 matrix by solving det(A - lambda I) = 0, yielding lambda -1 and 8, and an eigenvector for -1 such as (1,2).

Explore how neural networks, as a black box method, use artificial neurons to map input signals to outputs across diverse applications.

Explains how artificial neurons use weighted inputs, bias, and an activation function to produce outputs, and contrasts threshold, sigmoid, linear, tanh, and radial basis function with standardization and normalization.

Explore neural network topology, including input, hidden, and output layers, and learn how feedforward, recurrent, and backpropagation enable deep learning and function approximation.

Explore how neural networks adjust weights and biases through case and batch updates with backpropagation, learning rate, and stopping criteria to prevent overfitting.

Explore time series analysis and forecasting theory, focusing on autoregressive models and econometrics approaches like ARIMA, and distinguish univariate time series from cross-section data.

Explore common time series patterns, from upward trends and sine waves to random series, and identify the four components—trend, seasonality, cyclic, and randomness, guiding forecasts.

Explore univariate time series modeling to forecast future data using past values, including AR(p) and MA approaches with white noise assumptions and applications like inflation and stock prices.

Learn how moving average models rely on past error terms and white noise, and how ARMA and ARIMA integrate auto regression, moving average components, and differencing to achieve stationarity.

Infer ACF and PACF to decide how many past values matter for forecasting with AR, MA, or ARMA models, using correlograms and estimation methods.

Explore ARIMA forecasting for stock price indices by splitting data into training and testing sets. Fit ARIMA(1,1,1), evaluate coefficient confidence intervals and AIC, and select the best model for prediction.

Explore time series forecasting of stock prices using Prophet, addressing rising forecast error and random walk behavior, and learn data preparation with y and date parameters for Prophet.

Master time series data and deseasonalization in Excel by converting monthly airline passengers data to quarterly, using pivot tables to create clear, presentable reports.

Create an Excel line chart with markers for airline passenger data, label year and quarter on the horizontal axis, and apply four-quarter moving average CMA to reveal seasonality and trend.

Learn how to identify seasonality and irregular components in time series, deseasonalize data, and estimate trend with linear regression using Excel's data analysis tools.

Explore how gradient boosting machines create a sequential ensemble of weak models, weighting errors to optimize a differentiable loss and boost prediction accuracy.

Discover how gradient boosting turns weak classifiers into a strong learner, addresses overfitting with tricks, and blends h1, h2, h3 via weighted voting and bagging concepts.

Explore gradient boosting using decision tree stumps to partition data and reduce error rate. Learn how to weight weak learners with alphas and minimize error to create a wisdom-of-crowds ensemble.

Explore gradient boosting and AdaBoost as adaptive ensemble methods that improve weak learners by focusing on misclassified errors in a practical R example.

Explore how to structure a 12,000-record dataset for training and testing, build classification and regression trees with rpart, prune with cross-validation, and boost performance with gradient boosting.

Learn gradient boosting with gbm and adaBoost, tuning trees, shrinkage, and distribution options like bernoulli and multinomial to improve accuracy on training and test data.

Explore market basket analysis as an unsupervised learning method that discovers association rules among item sets using support, confidence, and lift, with differential analysis for retailer insights.

Interpret market basket rules by analyzing support, confidence, and lift with region-based transaction examples. Understand how x and y co-occur and influence probabilities.

Explore market basket analysis using a groceries data set from 30 days, learning about support, confidence level, and leap for association rules, and implementing with Excel Miner and R.

Explore market basket analysis in RStudio by inspecting a grocery csv, computing item frequencies, plotting top items, and deriving association rules with apriori, including support and confidence.

Explore association rules in Rstudio for market analysis by inspecting, sorting, and pruning rules based on support, confidence, and lift, and visualize with graph plots and interactive options.

Explore the data scientist's role in the big data landscape, including tools from ETL and Hadoop to visualization platforms, and Google's pre-trained machine learning APIs.

Explore how Google Vision API detects labels, faces, text via OCR, landmarks with longitude and latitude, logos, and explicit content, returning json for integration with Firebase, BigQuery, and NLP workflows.

Learn natural language processing with sentiment analysis and entity extraction, using Google BigQuery for analytics, the cloud natural language API, and RStudio workflows to analyze tweets and build word clouds.

The lecture demonstrates collecting and preprocessing Twitter data in R, converting tweets to text, creating and cleaning a corpus, generating a word cloud, and preparing for sentiment analysis.

Outlines the data science life cycle from problem definition to deployment, detailing data gathering, cleaning, feature engineering, modeling, evaluation, and algorithm selection with logistic regression and SVC.

Mastering artificial intelligence with Python and R introduces logistic regression for binary classification, comparing sklearn models like logistic regression, random forest, decision tree, and linear SVC.

Compute null value percentages and round them, then visualize survival patterns with seaborn plots and facet grids by age, sex, embarked port, and class.

The lecture demonstrates feature engineering: converting name to a numeric title, encoding sex and embarked, dropping the name column, and creating age group, fare group, and fare per person.

Draw the confusion matrix, run three-fold cross validation with logistic regression on the training data, and plot precision-recall and ROC curves while noting a ROC AUC of 97%.

Split data into training and test sets with caTools, then scale numeric features, apply binning (including weight of evidence) and try logistic regression to compare performance.

Fit a logistic regression model (glm, binomial) in R, inspect coefficients and p-values, predict on the test set, and use a stepwise, aic-based variable selection to improve accuracy near 84%.

Explore building lift charts and gain charts from a training set to evaluate predictive models, score customers by purchase probability, and compare train and test data.

Learn how to save and load models in R, use scores and deciles to evaluate performance, and deploy scoring with logistic regression for predictions.

Explore fitting a decision tree model and compare its performance to logistic regression. Learn data preparation, handling missing values, and training with rpart in R, noting scale not required.

Explore how to generate predictions with a classifier, view class probabilities by setting type=class, and assess model performance via confusion matrices, lift and gain charts, and pruning forests.