Machine Learning and Deep Learning Bootcamp in Python

Install Python from python.org on Windows, add Python to the path during setup, and verify environment variables; note that deep learning workflows may require Python 3.10 or the stable 3.8.

Install PyCharm as the Python IDE, set path and environment as needed, and run a sample project that prints hi PyCharm to confirm Python and IDE setup.

Install TensorFlow and Keras in PyCharm to power neural networks, deep learning, and reinforcement learning, using project settings to add libraries; Keras is an open source Python neural network library.

Explore why artificial intelligence and machine learning matter: optimize with heuristic and meta-heuristic methods, enable robotics and games, and unlock neural networks, deep learning, and reinforcement learning applications.

Explore the three main types of learning, supervised learning with labeled data, unsupervised learning for unlabeled data, and reinforcement learning with rewards and penalties, including Q-learning and deep Q-learning.

Explore fundamental statistics by computing the expected value (mean) of a random variable from outcomes and probabilities, and examine variance, standard deviation, covariance, and correlation, including their normalization.

Explore linear regression by modeling house prices with one or more explanatory variables. Train a linear model to predict price from size, using optimization to estimate b0 and b1.

Learn how to fit a linear regression model by minimizing mean squared error, compare design matrix and gradient descent approaches, and estimate parameters for house price prediction.

Explore gradient descent for linear regression, minimizing mean squared error with design matrices, updating b0 and b1 using partial derivatives and learning rate alpha to improve predictions.

Implement linear regression with numpy, pandas, matplotlib, and scikit-learn to predict house prices from square foot living, train the model with gradient descent, and visualize results.

Analyze linear regression results by computing mean squared error and r squared to assess model fit, analyze residual and total sums of squares, and discuss variance and predictions.

Explore logistic regression and maximum likelihood estimation, using sigmoid and log-likelihood to estimate parameters for binary outcomes. Learn gradient descent and Newton-Raphson methods with features like income and age.

Explore logistic regression on simple two-dimensional data, learning how the sigmoid function maps inputs to class probabilities and how gradient descent tunes b0 and b1 for binary classification.

Train a logistic regression model on the credit data using income, age, and loan features, then evaluate predictions on the test set with a confusion matrix and 91% accuracy.

Explore why logistic regression acts as a linear classifier with a linear decision boundary and a 0.5 sigmoid threshold in a two-feature setup.

Use five-fold cross validation with logistic regression on credit data to predict default from income, age, and loan, showing test scores and a mean accuracy near 91%.

Explore the k-nearest neighbor classifier, a supervised learning method that classifies by similarity using Euclidean and Manhattan distance. Use labeled data with features like sweetness and crunchiness, and majority vote.

Identify lazy learners, notably the k nearest neighbor classifier, which stores training data and makes slow predictions using Euclidean distance.

Learn how distance metrics power the k nearest neighbor classifier by measuring similarity with Euclidean and Manhattan distances, and when to use square Euclidean distance in higher dimensions.

Explore the bias-variance trade-off in machine learning, explain bias and variance, and show how to choose the optimal k for a k-nearest-neighbor classifier using cross-validation.

Explore k-nearest neighbor classification with a two-dimensional dataset of blue and red classes, using three neighbors, euclidean distance, and simple fit and predict steps.

Apply k-nearest neighbors to a credit dataset with income, age, and loan to predict defaults, using a 70/30 train/test split, min-max normalization, and confusion matrix evaluation to boost accuracy.

Explore how cross-validation selects the optimal k for a k-nearest neighbor classifier on credit scoring data, revealing k=32 with about 98% accuracy, using minimax transformation preprocessing.

Explore a concrete two-dimensional Naive Bayes classification example, computing prior and class-conditional probabilities within a neighborhood to classify a new point as green.

Explore the gaussian naive bayes classifier using the credit dataset, including training/test split, features like income, age, and loan, and evaluating with confusion matrix and 93% accuracy.

Learn text clustering by converting text to numerical features via tf-idf and tokenization, then group documents semantically or thematically, with Naive Bayes as a complementary approach.

Learn to apply tf-idf vectorization to create a document-term matrix and a similarity matrix, revealing which sentences are most alike in Python with scikit-learn.

Master text classification with naive bayes using the 20 newsgroups dataset, and build a text representation with countvectorizer and tf-idf transformer, followed by multinomial naive bayes.

Explore support vector machines, a powerful supervised classifier that finds the maximum-margin hyperplane to separate classes. Learn how support vectors define the model and how SVMs operate in infinite dimensions.

Learn how a support vector machine separates linearly separable data with a hyperplane w^T x + b = 0, maximizing the margin and solving the primal problem with Lagrange multipliers.

Learn how support vector machines manage non-linearly separable data by introducing slack variables and a tunable C parameter to balance misclassification and margin, and preview the kernel trick.

Explore kernel functions in support vector machines that transform data into a higher dimensional space to enable linear separation, using the kernel trick to compute inner products efficiently.

Import numpy, matplotlib, and scikit-learn to build a simple support vector machine, train on two-feature blue and red data, and plot the decision boundary and regions.

Classify the iris data set with a support vector classifier on four features across 150 samples and three classes, using a 70/30 split and evaluating with confusion matrix and accuracy.

Tune support vector machines on the iris data set using grid search cross-validation to explore kernels (linear, rbf, polynomial) and parameters c and gamma, identifying the best model with accuracy.

Apply support vector machines to classify handwritten digits using the digits dataset of 1797 eight by eight images, flattened into 64 pixel intensity features for input.

Apply a support vector classifier to handwritten digits, converting 8x8 images into 64 features and a 75/25 train test split. Evaluate with a confusion matrix and accuracy score, achieving 97%.

Explore the advantages and disadvantages of support vector machines, including memory-friendly use of support vectors for classification and regression, while noting slow performance, black box complexity, and no probability estimates.