At the end of this course, successful students will understand the logic behind and the high-level structure of several algorithms for Machine Learning tasks (e.g., estimators, preprocessing and scoring methods), will be able to implement them using common Python libraries (e.g., Pandas, NumPy, Matplotlib, Scikit-Learn), and will be able to produce interactive Jupyter notebooks to reproducibly disseminate details of their methods and results.  

Successful students will be able to perform  

• Exploratory Data Analysis (EDA) tasks using common Python libraries, including:  
  • Loading Data from flat CSV files containing structured, tabular datasets, including:  
    • local/remote file locations (e.g., local disk, OpenML.org)    
    • variables of different types (e.g., categorical, numerical)  
  • Summarizing Data, including:    
    • numerical and graphical descriptions (e.g., tables, figures)  
    • uni- and bi- variate descriptive statistics (e.g., centrality and dispersion measures, histograms, scatter plots)    
  • Preprocessing Data as appropriate, including:    
    • selecting specific columns/rows based on basic index or value criteria  
    • transforming (e.g., scale, center)  
    • factorizing (e.g., PCA for dimension reduction)  
• Supervised Machine Learning (ML) tasks using common Python libraries, including:  
  • Training models to predict classification/regression targets, including:
    • basic predictive estimators (e.g., kNN, Logistic Regression, linear SVM, Decision Tree)
    • non-linear extensions such as kernels or bagging/boosting ensembles (e.g., Random Forest, Stochastic Gradient Boost)
    • multi-layer perceptron estimators with varying network architectures (e.g., # of layers / neurons per layer) and back-propagation optimization parameter combinations (e.g., solver, epoch, learning rate)  
  • Selecting an optimally tuned model by assessing an individual estimators’ performance potential, relying on:  
    • a common data and measurement framework (e.g., k-fold cross-validation, fixed randomization seed)  
    • the exploration of several hyperparameter values (e.g., regularization)  
    • considerations about model generalization (e.g., bias/variance trade-off within measured performance results)    
    • Validating the selected model’s performance using scoring methods appropriate for the problem/question (e.g., precision, recall, F-1 score)  
• Unsupervised Machine Learning (ML) tasks using common Python libraries, including:  
  • Vector Quantization estimators (e.g., k-means) that partition the dataset into a fixed number of clusters  
  • Hierarchical Clustering estimators (e.g., agglomerative) that produce nested multi-level clusters (e.g., dendrograms) based on different linkages (e.g., single, complete)