The library rather than the statistics: the estimator API, pipelines, and evaluation done honestly.
Before this guide: Pandas, Deep Dive. For the statistics, see the Visual Handbook.
The notebooks
None of these notebooks is written yet. This is the plan, in reading order.
| 1 | The Estimator APIfit and predict, and why every model looks the same | not yet written |
| 2 | Your First ModelTrain, predict, score, in six lines | not yet written |
| 3 | Train and TestThe split, the seed, and what the test set is for | not yet written |
| 4 | PreprocessingScaling and encoding, and fitting on train only | not yet written |
| 5 | PipelinesWhat the pipeline exists to prevent, demonstrated | not yet written |
| 6 | Cross-Validationcross_val_score, the fold, and variance across folds | not yet written |
| 7 | Regression MetricsAnd what each single number hides | not yet written |
| 8 | Classification MetricsPrecision, recall, and the threshold as a business decision | not yet written |
| 9 | Imbalanced ClassesAccuracy that lies, and what to report instead | not yet written |
| 10 | Hyperparameter SearchGrid, random, and why nested CV exists | not yet written |
| 11 | Feature ImportancePermutation importance, and why the built-in one misleads | not yet written |
| 12 | Mixed Column TypesColumnTransformer, and one pipeline for a real table | not yet written |
| 13 | Saving a ModelPersisting, loading, and the version pin you will need | not yet written |
| 14 | A Complete WorkflowRaw table to a model you could defend in a meeting | not yet written |