Machine Learning with Python | Nesqual TechCourse outline
Framing machine learning problems
- From foundations to practiceLesson25 min
- Framing a prediction taskLesson30 min
- Meeting Kestrel Fibre and its dataLesson30 min
- The toolkit: plain Python and scikit-learnLesson30 min
- Success criteria before modellingLesson25 min
- Reproducible experimentsLesson30 min
- Chapter quiz: framing machine learning problemsQuiz15 min
Splits, leakage and baselines
Holding data back properly
Lesson30 min
Stratification: keeping the class balance in every splitLesson25 minTime-aware and group-aware splitsLesson30 minFinding and stopping leakageLesson35 minBaselines: the score to beatLesson30 minQuiz: splits, leakage and baselinesQuiz15 minLab: split the data, catch the leak, set the baselinesLab40 minPreparing features
- From rows to a design matrixLesson25 min
- Missing valuesLesson30 min
- Encoding categoriesLesson30 min
- Scaling numeric featuresLesson30 min
- Transforming and creating featuresLesson30 min
- The fit and transform discipline: building prep.pyLesson35 min
- Chapter quiz: preparing featuresQuiz15 min
Linear regression in practice
- Multiple regression and the normal equationsLesson35 min
- Reading coefficients without fooling yourselfLesson30 min
- Regression metrics and residualsLesson30 min
- Bends in the data: hinge and polynomial featuresLesson30 min
- Regularisation: ridge, and lasso in outlineLesson30 min
- The scikit-learn version and the one test evaluationLesson30 min
- Quiz: linear regression in practiceQuiz15 min
- Lab: an install-time model from scratchLab40 min
Logistic regression and probabilities
- Logistic regression from scratchLesson35 min
- Checking that training workedLesson30 min
- Regularisation: the L2 penaltyLesson30 min
- Reading the modelLesson30 min
- Class imbalance and class weightsLesson30 min
- The scikit-learn version and more than two classesLesson30 min
- Chapter quiz: logistic regression and probabilitiesQuiz15 min
Evaluating classifiers and choosing thresholds
- Why accuracy lies: the confusion matrixLesson30 min
- Precision, recall, F1 and specificityLesson30 min
- Choosing a thresholdLesson35 min
- Ranking quality: ROC, AUC and precision-recallLesson35 min
- Probability quality: log loss, Brier score and calibrationLesson35 min
- Uncertainty in a metric: bootstrap intervalsLesson35 min
- Quiz: evaluating classifiers and choosing thresholdsQuiz15 min
- Lab: evaluate the churn model and choose a thresholdLab40 min
Decision trees
- How a tree chooses a splitLesson30 min
- Growing a tree recursivelyLesson30 min
- Regression trees for install timesLesson30 min
- Controlling tree complexityLesson30 min
- Reading and explaining a treeLesson30 min
- Decision trees in scikit-learnLesson25 min
- Decision trees quizQuiz15 min
- Lab: grow and read a churn treeLab40 min
Ensembles: forests and boosting
- Why averaging helpsLesson30 min
- Random forests and out-of-bag estimatesLesson30 min
- Gradient boosting for regressionLesson35 min
- Boosting for classification and the libraries at workLesson30 min
- Comparing models fairlyLesson30 min
- Ensembles in scikit-learnLesson25 min
- Chapter quiz: forests and boostingQuiz15 min
Nearest neighbours and clustering
- k-nearest neighbours in practiceLesson30 min
- Distances and what they costLesson30 min
- k-means properlyLesson35 min
- Choosing the number of clustersLesson30 min
- Describing and using segmentsLesson30 min
- Beyond k-means: DBSCAN, hierarchical clustering and PCALesson25 min
- Quiz: nearest neighbours and clusteringQuiz15 min
- Lab: neighbours and segments for KestrelLab40 min
Cross-validation and tuning
- One validation split is noisyLesson30 min
- Cross-validation without leakingLesson35 min
- Hyperparameter searchLesson30 min
- Not fooling yourselfLesson30 min
- Learning curves and validation curvesLesson30 min
- Putting it together: an honest tuning reportLesson30 min
- Cross-validation and tuning quizQuiz15 min
- Lab: tune a churn tree with honest cross-validationLab40 min
Error analysis, interpretation and fairness
- Error analysis: reading the mistakesLesson35 min
- What the model relies on: global interpretationLesson35 min
- Explaining one predictionLesson30 min
- Measuring fairness across groupsLesson35 min
- Model cards and responsible useLesson30 min
- Error analysis, interpretation and fairness quizQuiz15 min
Capstone: the Kestrel churn model
- The capstone briefLesson25 min
- Planning: audit, baselines and a budgetLesson30 min
- Worked walkthrough: logistic regression chosen by CVLesson30 min
- Finishing: threshold, refit and one testLesson35 min
- Saving the model, the model card and a final reviewLesson30 min
- The capstone in scikit-learn, and where nextLesson25 min
- Chapter quiz: the Kestrel churn capstoneQuiz15 min
- Lab: deliver the Kestrel churn modelLab40 min