Data-Centric Machine Learning with Python: Hands-On Guide

Discover why Python is the go-to language for data science and how to use NumPy, pandas, matplotlib, Seaborn, and scikit-learn across data collection, cleaning, exploration, modeling, evaluation, and deployment.

Learn pandas basics for data manipulation with series and data frames, loading data, exploring with head, tail, sample, descriptive statistics, and performing merges, joins, and groupby aggregations.

Explore NumPy for numerical computing, including 1D, 2D, and 3D arrays, broadcasting, and universal functions, and learn to create, reshape, slice, and perform element-wise operations.

Explore data visualization with matplotlib and seaborn to reveal patterns using line, bar, histogram, scatter, and box plots. Master subplots, heatmaps, and pair plots with seaborn datasets like tips.

Explore exploratory data analysis (EDA) to analyze and visualize data sets, uncover patterns, detect outliers, identify missing values, test hypotheses, and understand data structure for cleaning and feature selection.

Discover how to detect, understand, and handle missing values using dropna, imputation (mean, median, mode), forward/backward fill, interpolation, kNN, and MICE, with Titanic data and model evaluation.

Identify and handle outliers using box plots, scatter plots, z scores, and IQR, then apply removal, clipping, or median imputation to improve model performance.

Apply feature engineering to create and modify features, including transforming variables, encoding categorical features (one hot encoding and label encoding), creating interaction features, scaling to boost accuracy and reduce overfitting.

Apply feature selection to improve model performance and interpretability by choosing relevant features. Utilize variance threshold, correlation threshold, RFE, L1 regularization, and tree-based methods like random forest and gradient boosting.

Explore encoding categorical features for machine learning, converting nominal and ordinal data into numerical formats using one-hot, label, ordinal, binary, frequency, and target encoding.

Explore dimensionality reduction techniques such as PCA, LDA, and t-SNE on the Titanic dataset to simplify models, reduce computation, cut overfitting, and enhance visualization and interpretability.

Balance imbalanced datasets using random oversampling, undersampling, Smote, and balanced random forest. Assess performance with precision, recall, F1 score, and ROC AUC, and apply cost-sensitive learning via class weights.

Explore regression as a core supervised learning method to predict continuous outcomes using simple, multiple, linear, and non-linear models, including polynomial regression, with applications in finance, health, and house prices.

Explore linear regression, including simple and multiple regression, assumptions such as linearity, independence, homoscedasticity, and normality, and evaluate with r-squared and mean squared error.

Explore ridge (L2) and lasso (L1) regression to curb overfitting with regularization, shrink coefficients, and perform feature selection; compare elastic net and assess model performance via mean squared error.

Explore regression evaluation metrics, including mean absolute error, mean squared error, RMSE, R-squared, and adjusted R-squared, with Titanic dataset examples and guidance on when to use each.

Explore logistic regression for binary classification in data-centric machine learning with Python, using the sigmoid function to estimate probabilities. Evaluate with accuracy, precision, recall, F1, and ROC AUC on Titanic.

Explore decision trees as intuitive, interpretable models for classification and regression, with splitting criteria like Gini index, entropy, and information gain, and pruning to prevent overfitting.

Explore decision trees for classification and regression, including splitting criteria like Gini index, entropy, and information gain; learn pruning and evaluation with cross-validation, accuracy, mean squared error, and confusion metrics.

Explore random forests, an ensemble of decision trees using bootstrap sampling and random feature selection to boost accuracy and reduce overfitting, and learn evaluation metrics like accuracy and F1 score.

Explore evaluation metrics for classification, including accuracy, precision, recall, F1 score, ROC and AUC, and use the confusion matrix to assess model performance beyond accuracy on imbalanced data.

Master train-test split to evaluate model generalization on unseen data with 80/20 split and random_state, while guarding against data leakage and using metrics like accuracy, precision, recall, F1, and ROC-AUC.

Explore cross-validation techniques to assess model performance and prevent overfitting, including k-fold, stratified k-fold, leave-one-out, and repeated k-fold, with hyperparameter tuning in mind.

Learn how to tune hyperparameters for machine learning models using grid search and randomized search in Python, with cross-validation to prevent overfitting and optimize performance.

Learn to evaluate models for generalization using classification and regression metrics. Use accuracy, precision, recall, F1, confusion matrix, ROC-AUC, MAE, MSE, RMSE, and R-squared, plus cross-validation and train-test-validation splits.

Explore unsupervised learning through clustering with k-means, hierarchical clustering, and DBSCAN on iris dataset, enabling exploratory data analysis and anomaly detection, evaluate clusters with silhouette score and elbow method.

Learn k-means clustering, an efficient unsupervised method that partitions data into k clusters around centroids, with Python on the iris dataset, including standardization, elbow and silhouette methods, and k-means++ initialization.

Explore hierarchical clustering, including agglomerative and divisive approaches, and visualize clusters with dendrograms. Learn to standardize data, evaluate with silhouette score, and choose linkage methods to minimize within-cluster variance.

Learn anomaly detection with isolation forests, an ensemble method using random partitions to isolate anomalies and compute anomaly scores, with domain knowledge and contamination-aware feature scaling.

Learn to preprocess text data with tokenization, stopword removal, lowercasing, and punctuation stripping, then vectorize using bag of words and tf-idf for a logistic regression sentiment model.

Build an image classification model with a convolutional neural network using Fashion-MNIST from Keras, including loading, normalizing, and splitting data. Use data augmentation, transfer learning, and regularization to improve generalization.

Explore recommendation systems using item-based collaborative filtering with cosine similarity on Movielens 100k data, building a user-item matrix and top similar-movie recommendations, and evaluating with precision and recall.

Explore ensemble methods that combine multiple models to boost accuracy and robustness for classification and regression, including bagging, boosting, voting, and stacking, with examples like random forest and XGBoost.