Süni İntellekt Mühəndisliyi (AI Engineer)

Kurs Haqqında

• Understanding the Problem & Inspecting Raw Data
  • 1 Framing the Business / Research Question
  • 2 Clarifying objectives
  • 3 Defining target variable
  • 4 Success metrics
  • 5 Importing Data Sources
  • 6 Flat files (CSV, Excel)
  • 7 SQL queries
  • 8 APIs
  • 9 First-Look Inspection in Pandas
  • 10 head(), info(), describe(), data types, memory usage
  • 11 Define a clear business or research question
  • 12 Data Structure Concepts
  • 13 Acquire and clean the data, then explore it to understand its patterns
  • 14 Rectangular vs. nested data
  • 15 Engineer features and select appropriate models from those studied (e.g. linear models, trees, boosting, SVMs, etc.)
  • 16 Tidy-data principles
  • 17 Train, tune, and evaluate their models using best practices (cross-validation, learning curves, proper metrics)
  • 18 Indexes
  • 19 Package their workflow into a reproducible pipeline and deliver a concise report summarizing their approach, findings, and lessons learned
  • 20 Quick Data Quality Scan
  • 21 Missing-value counts
  • 22 Duplicated rows
  • 23 Identifying impossible values
• Project Delievery and Interview
• Handling Missing Data & Outliers
  • 1 Diagnosing Missingness
  • 2 MCAR, MAR, MNAR
  • 3 Visual tools (missingno, heatmaps)
  • 4 Imputation Strategies
  • 5 Mean / median
  • 6 Constant
  • 7 Flag columns
  • 8 Outlier Detection
  • 9 Z-score
  • 10 IQR rule
  • 11 Tukey fences
  • 12 Domain-based limits
  • 13 Robust Rescaling
  • 14 Winsorization
  • 15 log / Box-Cox transforms to tame skew
  • 16 Practical Guidelines
  • 17 When to drop vs. impute
  • 18 Documenting every assumption
• Exploring Data Characteristics
  • 1 Descriptive Statistics Deep-Dive
  • 2 Mean, median, mode, trimmed mean
  • 3 MAD
  • 4 CV
  • 5 Distribution Analysis
  • 6 Histograms, KDE, QQ-plots
  • 7 Detecting skew & kurtosis
  • 8 Visual Exploration of Categorical Data
  • 9 Frequency tables
  • 10 Bar charts
  • 11 Pareto plots
  • 12 Relationships in Numeric Data
  • 13 Scatter-matrix
  • 14 Correlation heatmap
  • 15 Pair-plot diagnostics
  • 16 Initial Feature Insights
  • 17 Signals
  • 18 Redundancies, and surprises worth engineering later
• Data Transformation & Feature Preparation
  • 1 Encoding Categorical Variables
  • 2 One-hot and Ordinal
  • 3 Target & frequency encoding
  • 4 Pitfalls of high-cardinality
  • 5 Scaling & Normalization
  • 6 StandardScaler, Min-Max, RobustScaler
  • 7 Choosing per algorithm
  • 8 Feature Generation Basics
  • 9 Date-time parts, qggregation, ratios, interaction terms
  • 10 Pipeline Construction in scikit-learn
  • 11 ColumnTransformer
  • 12 Custom transformers
  • 13 Reproducible workflows
• (Lab). EDA & Cleaning
  • 1 Load a real-world dataset with mixed types
  • 2 Perform systematic missing-value treatment and outlier fixes
  • 3 Produce distribution and relationship visualizations
  • 4 Build a preprocessing pipeline that outputs a model-ready table
  • 5 Present a short notebook summarizing insights and cleaned data quality metrics
• Machine Learning
  • 1 Linear Models I - Simple & Multiple Regression
  • 2 Regression Fundamentals
  • 3 Hypothesis & ordinary least squares (OLS)
  • 4 Cost computation: MSE & Normal equation
  • 5 Multiple Regression
  • 6 Vectorized implementation
  • 7 Impact of feature scaling on solvers
  • 8 Model Evaluation
  • 9 Metrics: MSE, RMSE, R2
  • 10 Cross-validation overview
• Linear Models II - Polynomial & Interaction
  • 1 Polynomial Regression
  • 2 Feature expansion with PolynomialFeatures
  • 3 Selecting polynomial degree
  • 4 Effect on model capacity
  • 5 Bias–Variance Trade-Off
  • 6 Underfitting vs. overfitting symptoms
  • 7 Impact of model complexity
  • 8 Learning Curves
  • 9 Plot training vs. validation error
  • 10 Diagnosing capacity issues
• Regularized Regression — Ridge, Lasso, Elastic Net
  • 1 Why Regularize?
  • 2 Over-fitting symptom review
  • 3 Shrinkage concept & coefficient paths
  • 4 Ridge Regression (L2)
  • 5 Cost function with λ |w|²
  • 6 Effect on correlated features
  • 7 Closed-form vs. SGD implementation
  • 8 Lasso Regression (L1)
  • 9 Sparse solutions & feature selection
  • 10 Geometry: diamond constraint region
  • 11 When to prefer over Ridge
  • 12 Elastic Net
  • 13 Mixing L1 & L2 penalties
  • 14 Tuning α and l1_ratio with grid-search CV
• Logistic Regression
  • 1 Logistic Regression
  • 2 Sigmoid activation & probability estimates
  • 3 Cross-entropy loss & gradient descent
  • 4 Naïve Bayes
  • 5 Bayes’ theorem with feature-independence
  • 6 Gaussian vs. Multinomial variants
  • 7 Model Evaluation
  • 8 ROC curve & AUC
  • 9 Precision, recall & F1-score
• Regression & Classification Lab
  • 1 Hands-On Regression
  • 2 Fit linear & polynomial models
  • 3 Experiment with ridge & lasso regularization
  • 4 Hands-On Classification
  • 5 Train logistic & Naïve Bayes classifiers
  • 6 Evaluate with ROC & precision/recall
• K-Nearest Neighbors
  • KNN Fundamentals
  • Distance metrics & choice of k
  • Uniform vs. distance-weighted voting
  • Impact of feature scaling
  • High-Dimensional Data
  • Curse of dimensionality
  • Nearest-neighbor search strategies
  • Dimensionality reduction pre-processing
  • Model Evaluation
  • Cross-validation for k selection
  • Confusion matrix & classification report
  • Maximal Margin Classifiers
    • 1 Hard vs. soft margin formulation
• Support Vector Machines
  • 1 Slack variables & C parameter
  • 2 Kernel Trick
  • 3 Linear, polynomial & RBF kernels
  • 4 Implicit feature mapping
  • 5 Model Tuning
  • 6 C & γ trade-offs
  • 7 Grid search & scaling requirements
• SVM Project
  • 1 Hyperparameter Tuning
  • 2 GridSearchCV over C ([0.1,1,10]) and gamma ([0.01,0.1,1]).
  • 3 Retrain best model and report test accuracy.
  • 4 Data Prep
  • 5 Load a two-class dataset (e.g. make_moons).
  • 6 Split train/test and apply StandardScaler.
  • 7 Linear SVM
  • 8 Train LinearSVC.
  • 9 Evaluate accuracy, confusion matrix, classification report.
  • 10 Plot decision boundary.
  • 11 Kernel SVM
  • 12 Train SVC(kernel='rbf') and SVC(kernel='poly', degree=3).
  • 13 Evaluate & plot each.
• Decision Trees
  • 1 Decision Trees (title listed in PDF; detailed subtopics not specified on the same page).
• Random Forests
  • 1 Bagging & Variance Reduction
  • 2 Bootstrap sampling
  • 3 Aggregating predictions
  • 4 Random Subspace Method
  • 5 Random feature selection per split
  • 6 Decorrelated trees
  • 7 Out-of-Bag Evaluation
  • 8 OOB error estimation
  • 9 Feature importance from OOB samples
  • 10 Tree Construction
  • 11 Splitting criteria: Gini vs. entropy
  • 12 Recursive binary splitting
  • 13 Tree Regularization
  • 14 Pre-pruning: max_depth, min_samples
  • 15 Post-pruning techniques
  • 16 Interpretability
  • 17 Feature importance
  • 18 Tree visualization
• Tree Ensemble Methods Lab
  • 1 Decision Tree Practice
  • 2 Train & visualize a single tree
  • 3 Experiment with pruning parameters
  • 4 Random Forest Practice
  • 5 Fit RF model & measure OOB error
  • 6 Compute & plot feature importances
  • 7 Hyperparameter Tuning
  • 8 Grid search over n_estimators & max_features
• Boosting
  • 1 Boosting Concepts
  • 2 AdaBoost: weighted resampling & exponential loss
  • 3 Gradient Boosting: additive trees & loss-gradient descent
  • 4 Key Hyperparameters
  • 5 n_estimators, learning_rate & max_depth
  • 6 Shrinkage vs. over-fitting trade-off
  • 7 XGBoost
  • 8 Tree regularization (L1/L2) & pruning
  • 9 Column subsampling & tree learning
  • 10 LightGBM
  • 11 Histogram-based binning & leaf-wise growth
  • 12 GPU & multi-threaded acceleration
  • 13 CatBoost
  • 14 Ordered boosting & target leakage prevention
  • 15 Built-in categorical feature handling
• Anomaly Detection
  • 1 Isolation Forest
  • 2 Random partitioning principle
  • 3 Anomaly score via path length
  • 4 Key hyperparams: n_estimators, max_samples
  • 5 One-Class SVM
  • 6 Kernel-based novelty detection
  • 7 ν parameter & decision boundary
  • 8 Scale sensitivity
• Boosting & Anomaly Detection Lab
  • 1 Gradient Boosting Practice
  • 2 Train & tune XGBoost and LightGBM
  • 3 Compare learning curves
  • 4 Anomaly Detection Practice
  • 5 Fit Isolation Forest & One-Class SVM
  • 6 Evaluate with ROC & precision-recall metrics
• Clustering — K-Means & Hierarchical
  • 1 K-Means Clustering
  • 2 Centroid initialization & update steps
  • 3 Choosing k via elbow & silhouette methods
  • 4 Hierarchical Clustering
  • 5 Agglomerative vs. divisive approaches
  • 6 Linkage criteria: single, complete, average
  • 7 Dendrogram Interpretation
  • 8 Cutting trees for cluster assignment
• Dimensionality Reduction — PCA
  • 1 Principal Component Analysis
  • 2 Covariance matrix & eigen decomposition
  • 3 Explained variance ratio
  • 4 Dimensionality Reduction Workflow
  • 5 Centering & scaling data
  • 6 Selecting number of components
  • 7 Applications & Visualization
  • 8 Preprocessing for ML & 2D/3D plotting
• Recommendation Systems
  • 1 Introduction
  • 2 What & why they matter
  • 3 Key use cases (e-commerce, streaming, news)
  • 4 Collaborative Filtering
  • 5 Memory-based: user–user & item–item similarity
  • 6 Model-based: matrix factorization (SVD/ALS)
  • 7 Content-Based Filtering
  • 8 Item profiles (metadata) & user profiles
  • 9 Similarity via TF-IDF or embeddings
  • 10 Hybrid Systems
  • 11 Weighted & switching strategies
  • 12 Example: Netflix’s metadata + viewing history
• Neural Networks and Deep Learning (AI)
  • 1 What is Artificial Intelligence?
  • 2 Differences between AI, ML, and Deep Learning
  • 3 Modern AI systems and their capabilities
  • 4 Overview of Neural Networks
  • 5 Biological inspiration and historical development
  • 6 Discussion: "Can machines really think?"