Data Science for AI and Machine Learning Using Python

Explore the foundations of statistics, machine learning, and data science with hands-on Python practice, covering data types, distributions, and data engineering to transform and model data.

Set up a python 3.x environment with anaconda and spyder, import numpy and pandas, configure the working directory and util, apply ggplot style, and use a seed for reproducible results.

Master pandas dataframe basics by loading the mpg dataset, inspecting data types and shapes, converting column names to uppercase, and identifying numeric, categorical, and target variables.

Convert the origin column to a category in the pandas dataframe and review basic statistics, correlations (mpg and cylinders, negatively correlated), and kurtosis with its distribution meaning.

Learn pandas data frame column operations by accessing data with iloc and loc, selecting rows and columns, computing counts and minimums, handling NaN, and filtering with conditions.

learn panda operations to create a temp_col from mpg divided by cylinders, drop columns, reset index, then use np.where and string methods like lower and contains.

Learn to merge and concatenate data frames in pandas, combining salary and HR data and performing row-wise and column-wise joins, including left, right, and outer joins.

Learn to generate and read summary tables, compute mean, count, and sums by category, and add min and max statistics to prepare business-ready reports for end users.

Explore graphs in Python to visualize data relationships and distributions, using scatter plots of horsepower vs. mpg, histograms, and box plots, including outlier interpretation.

Master numpy basics by converting lists to arrays, inspecting shape, and indexing and slicing 1D and 2D data, including negative indices, to prepare for data science and deep learning.

Explore slicing two dimensional data with NumPy, performing horizontal and vertical splits into train and test sets, and reshaping data for deep learning workflows.

Explore two-dimensional stacking with numpy and pandas joins, learning horizontal and vertical stacking, array appends, and memory concepts like shallow versus deep copy.

Learn how to generate and use a data dictionary that describes each feature with basic statistics, data type, missing values, and remarks for scalable machine-learning projects.

Explore descriptive analysis of numeric variables with graphical methods, using a function to generate histograms and density plots to visualize distributions.

Learn to perform descriptive analytics on two numeric variables—engine displacement and acceleration—and their relationship to the response, using scatterplots and boxplots to spot patterns and outliers.

Explore categorical and numeric descriptive analysis by examining origin categories, their counts and mean mpg per category, while assessing skewness, correlations, and box plots with Python tools.

Explore supervised machine learning with a focus on regression, and master preprocessing: handling missing data, transforming features, feature engineering and interaction, plus checking collinearity.

Explore feature selection techniques—forward selection, backward elimination, and recursive feature elimination—along with model building and regularisation using L1 (lasso) and L2 (ridge) penalties.

Explore residual analysis to assess regression models by comparing actual versus predicted values, examining autocorrelation with Durbin-Watson, variance behavior (homoscedasticity), outliers, and normality via the Q-Q plot.

Begin hands-on regression by loading the data, computing MSE, RMSE, and MAE, and preparing a dataset with categorical columns and robust validation.

Assess normality with histograms, Q-Q plots, and statistical tests, then apply Box-Cox transformation to normalize skewed data, and verify by inverse transformation and comparing original and transformed data.

Build a simple linear regression model in Python, split data into train and test sets, train the model, and evaluate with MAE, RMSE, and R-squared while visualizing residuals.

Explore linear regression on numeric features by preparing data, dropping categorical columns, and evaluating predictions with MAE, RMSE, and R-squared, plus residual analysis and Durbin-Watson checks.

Learn how hot encoding converts categorical columns into numeric dummy variables and how scaling normalizes features to a common range for regression.

Apply linear regression to hot encoded and scaled data, evaluate MAE, RMSE, and R squared, and observe modest improvements over prior results while noting normality concerns in residuals.

Explore how tree structures drive classification and regression, and how features are selected by information gain, entropy, and gini to split into leaves.

Learn CatBoost, an open-source boosting algorithm from Yandex that handles categorical features without hot encoding, offers good performance with small data, and integrates with TensorFlow in Python.

Discover how to tune CatBoost hyperparameters to minimize RMSE, using manual search, grid search, and random grid search with practical guidelines on combinations and run times.

Explore XGBoost, an extreme gradient boosting method for regression that builds sequential weak trees on data subsets and combines them into a strong predictive ensemble via weighted averages.

Apply hands-on XGBoost using DMatrix, install xgboost, prepare data, split train-test, and tune core parameters like eval_metric, eta, col_sample, max_depth, and nthread.

Learn how to train an xgboost model with xgb.train and DMatrix, tune the number of rounds, evaluate with MAE, plot feature importance, and predict on test data.

Learn classification, where outcomes are nominal or ordinal rather than numeric, and use confusion matrices, CART, and metrics like precision, recall, and ROC AUC.

Engage in hands-on classification by loading the iris dataset, targeting species as the categorical label, and applying a random forest with train-test split.

Explore how random forest uses bagging and bootstrap samples to build multiple decision trees in parallel and ensemble their results for classification and regression.

Explore random forest classification in Python by tuning key parameters like n_estimator and min samples per leaf, training a classifier, and evaluating results with confusion matrices, accuracy, and probabilities.

Train a catboost classifier for a multi-class task by mapping the target to integers and evaluating results with a confusion matrix and metrics such as precision, sensitivity, and specificity.

Use one-class SVM for anomaly detection in highly imbalanced data, such as fraud analytics, training on a single class and evaluating with accuracy and sensitivity.

Explore logistic regression as a binary classification technique that predicts 0 or 1 and yields a probability between 0 and 1, with applications such as fraud detection.

Explore hands-on logistic regression for binary classification and compare it with random forest, catboost, and xgboost using practical preprocessing, train-test split, and confusion matrix evaluation.

Explore bias vs. variance to assess model goodness, distinguishing underfit and overfit through train and test errors, and apply more data, features, regularization, or ensembles to improve performance.

Explore unsupervised learning with clustering, PCA, and KNN, and learn how to determine cluster numbers using silhouette and elbow methods, with practical intuition.

Explore unsupervised clustering with k-means using iris data, including scaling, elbow method for choosing k, and PCA for 2D visualization, then map clusters to species and interpret attributes.

Explore agglomerative clustering, including linkage options such as ward, average, complete, and cosine, with PCA visualization and three clusters.

Learn how k-nearest neighbors are calculated with the NearestNeighbors library and inspect distances. Apply the method to simple data and iris data, and visualize results with a scatter plot.

Apply k-nearest neighbors to iris data by computing distances to locate neighbors. Visualize on PCA or original data, test with samples, and note kNN’s use in regression and classification.

Explore supervised and unsupervised analytics, including regression and clustering. Apply to customer segmentation, targeted marketing, and recommendation systems, with dimensionality reduction via PCA for visualization and feature elicitation.