Machine Learning with Python

Explore toy datasets to practice supervised learning tasks, including classification and regression, and examine unsupervised and reinforcement learning concepts using real and example datasets.

Explore the iris dataset using the package to inspect its structure, feature names, target labels, and the balanced three-class distribution for supervised learning.

Explore remaining toy datasets in Python machine learning, inspect wine dataset attributes such as feature names, target names, class distribution, and data statistics to guide practical data exploration.

Create a data frame from a sklearn Bunch by loading the iris data and mapping feature names to columns. Convert the target to labels and inspect the shape for modeling.

Create a bunch with your own data by importing data from files or databases, shaping the data, and preparing a dataset for machine learning experiments.

Create a data bunch from your own data by defining target, features, and descriptions; manage file names, keys, and Excel attachments to build a dataset for machine learning in Python.

Learn the k nearest neighbors algorithm, derive distance measures like Euclidean and Manhattan, compare feature vectors for known and unknown samples, and understand the role of the p parameter.

Explore how to classify an unknown sample using euclidean distance to known samples and nearest neighbors. See how height weight features inform distance calculations and predict good or bad labels.

Apply a k-nearest neighbors approach in Excel to predict an unknown flower's name from known flowers using Euclidean distance and labeled features. Tune the number of neighbors to optimize accuracy.

Explore how the k-nearest neighbors parameter n_neighbors determines the label of an unknown sample, examining 1, 2, 3, and 5 neighbors, tie scenarios, and the role of parameter tuning.

Explore Hamming distance as a measure of mismatches between strings, compare it with Manhattan distance, and apply Python coding to problems and game-like scenarios.

Implement a k-nearest neighbors estimator from scratch in Python by computing Euclidean distances between samples and predicting unknowns using the nearest neighbors, guided by a driver program.

Identify the most similar neighbors by computing distances between each unknown sample and known samples, then sort and select the top k neighbors to guide prediction.

This lecture demonstrates writing code to classify new samples by finding nearest neighbors with Euclidean distance, predicting labels for moon samples, and mapping results to the user interface.

Gather data and define X and y, then preprocess, split into training and testing sets, train and evaluate, and deploy via an API with testing tools.

Load and inspect a csv file, select four feature columns as X and the target column as y, and encode the categorical labels into numeric codes for modeling.

Learn to split data into training and testing sets using a model selection tool, control with train/test sizes, shuffle, random_state, and stratify, and unpack features and labels for evaluation.

Learn how to train and fit a model in machine learning with Python, using a nearest neighbors classifier, including selecting parameters and evaluating on training data.

Learn to split data into training and testing, train a model, and predict labels for test data, using predicted probabilities and accuracy to evaluate performance.

Explore calculating model accuracy with accuracy score and a confusion matrix by comparing predicted labels to real data labels, and see how stratified sampling and random state affect results.

Explore practical hyperparameter tuning for k-nearest neighbors using Python, comparing parameter choices like number of neighbors and distance metrics, and employing grid search cross-validation to optimize accuracy.

Discover cross validation and grid search for hyperparameter tuning, using training and testing data to estimate model accuracy, including stratified cross validation.

Explore grid search cross-validation for hyperparameter tuning in machine learning with Python, build a parameter grid as a list of dictionaries, and evaluate with three-fold cross-validation to select best parameters.

Explore randomized search cross-validation for hyperparameter tuning, assess parameter options, and fit the model with the best configuration observed to improve performance.

select optimal parameters, train with the best number of neighbors, then save the trained model using pickle’s dump function, storing it in binary format for later loading and production.

Learn to save and load machine learning models in binary format using pickle in Python, load the model into an object, and perform predictions with prepared data.

Explore practical machine learning workflows in Python, including working with nearest neighbors, parametric and non-parametric methods, distance metrics like Manhattan and Euclidean, and cross-validation for model tuning.

Explore the mathematics of linear regression, deriving the y = mx + b model, understanding slope and intercept, and applying it to continuous outcomes while noting regression assumptions.

Learn to predict a baby's weight from age in months using linear regression, estimate the intercept and slope with ordinary least squares, and analyze data relationships and correlation.

Explore ordinary least squares for simple linear regression: derive weights b0 and b1 from two-column data with months and rate, using X transpose and matrix forms to predict rate.

Estimate parameters using ordinary least squares to fit a linear regression with two or more features, determining the intercept and coefficients to map inputs X to outputs y.

Learn to find parameters using sklearn by fitting a linear regression model, interpreting the intercept and coefficients, and handling features, including dummy features, with practical steps.

This lecture explains simple linear regression with one variable, deriving b0 and b1 from covariance and variance using means and sums to calculate the parameters.

Learn to build a multivariate linear regression model to predict baby weight from age, month, and gender, estimating intercept and coefficients for multiple inputs.

Learn how to use linear regression to predict lifespan from features, compare single- and multi-variable models, prepare data, and assess predictions with metrics such as accuracy.

Apply linear regression to predict life span from fertility features, fit the model, interpret the intercept and coefficients, and predict on new data with visualization.

Explore evaluating regression models with r-squared, sse, sst, and mean squared and mean absolute metrics, understanding that higher r-squared indicates a better fit, with Excel and Python demonstrations.

Compare supervised regression metrics, focusing on rmse and mse for evaluation. Demonstrate calculating mse and rmse from actual versus predicted values, and mention mean absolute error as an alternative.

Build a linear regression model to predict life span from eight input features, including population, fertility, HIV, carbon dioxide, and GDP, and evaluate with train-test splits and r-squared metrics.

Explore cross validation methods for linear regression, including fivefold and k-fold, to train, validate, and test models. Learn to interpret cross validation scores to estimate accuracy on unknown data.

Explore linear regression on the Boston housing dataset, using multiple features to predict price, fit the model, evaluate accuracy and standard deviation, and review dataset documentation.

Explore the logistic regression approach for binary diabetes prediction on the Pima Indians dataset, using the sigmoid function, train-test split, and accuracy metrics.

Explore logistic regression math: from linear regression foundations to the sigmoid function, intercepts, and coefficients for binary classification; examine practical intuition with hands-on equations and example scenarios.

Apply logistic regression to predict diabetes using the Pima Indians data, using sigmoid outputs and accuracy metrics, with train-test split and evaluation via confusion matrix and classification report.

Explore how a confusion matrix and confusion metrics quantify correct and wrong predictions, and how accuracy is derived from real versus predicted values across positive and negative cases.

Explore the classification report to interpret per-label accuracy, precision, recall, and f1 score through the confusion matrix, including macro and weighted averages for unbalanced data.

Explore the receiver operating characteristic curve to visualize classifier performance by plotting false positive rate versus true positive rate (sensitivity) and evaluating thresholds and the area under the curve.

Compute area under the curve via integration and apply it to AUC evaluation with a parabola example. Review metrics for classification and regression, including accuracy, confusion matrix, and ROC-based AUC.

Explore support vector machines for classification, emphasizing a clear margin of separation, high-dimensional processing, and memory-efficient use of support vectors.

Explore support vectors and the margin maximization in a linear svm, using synthetic two-class data, and learn how support vectors determine the decision boundary with a linear kernel.

Explore how to identify linear vs non-linear data, use a linear kernel for linear data, and switch to an rbf kernel to separate circular patterns with a clear margin.

Apply support vector classification to the iris toy dataset using two features, train a linear model, and achieve about 0.8 accuracy; use grid search to tune C and gamma.

Use a pipeline with PCA-based feature engineering and an SVC classifier to perform face recognition on labeled faces in the wild dataset, with hyperparameter tuning and evaluation via confusion matrix.

Identify and remove outliers in data preprocessing using a box plot with Q1 and Q3, iqr, and fences. Learn practical steps for applying these bounds during data engineering preprocessing.

Identify four outliers among 96 items with mean and standard deviation, delete them to yield 92 observations, and verify ten columns have no outliers using a box plot for predictions.

Explore preprocessing categorical features in Python machine learning, converting string data to numeric, encoding categories, and selecting features such as reason and country to improve lifespan prediction with linear regression.

Convert categorical features to numeric using pandas get_dummies, handling multiple distinct values and optional drop_first, and relate it to one-hot encoding in sklearn and keras.

Apply ridge regression to datasets with categorical features, including region, by converting categories to numeric, and compare cross-validated accuracy with linear models.

Learn how to handle missing data by converting values to nan, dropping rows with missing values, or imputing with mean, median, or most frequent values for numeric and string columns.

Learn to build a pipeline that unites pre-processing, imputation and feature engineering with SelectKBest and ANOVA, and a SVC estimator, including memory caching.

Construct a machine learning pipeline that applies a simple imputer for missing values with the most frequent strategy, then fits an svc estimator with a train-test split.

Explore how entropy and Gini serve as criteria in decision tree classifiers, with practical intuition from bucket examples, probability, and entropy calculations.

Compute entropy on a five-record weather dataset, evaluating outlook, humidity, wind, and temperature, to build a decision tree and identify top features.

Explore entropy calculation and the Gini index for building decision trees using a weather dataset, highlighting how to compute probabilities, feature splits like outlook, temperature, humidity, and wind.