Examples#
Forecasting#
- Forecasting with sktime
- Table of Contents
- 1. Basic forecasting workflows
- Step 1 - Preparation of the data
- Step 2 - Specifying the forecasting horizon
- Step 3 - Specifying the forecasting algorithm
- Step 4 - Fitting the forecaster to the seen data
- Step 5 - Requesting forecasts
- 1.2.1 The basic deployment workflow in a nutshell
- 1.2.2 Forecasters that require the horizon already in
fit
- 1.2.3 Forecasters that can make use of exogeneous data
- 1.2.4. Multivariate forecasting
- 1.2.5 Probabilistic forecasting: prediction intervals, quantile, variance, and distributional forecasts
- 1.2.6 Panel forecasts and hierarchical forecasts
- Step 1 - Splitting a historical data set in to a temporal train and test batch
- Step 2 - Making forecasts for y_test from y_train
- Steps 3 and 4 - Specifying a forecasting metric, evaluating on the test set
- Step 5 - Testing performance against benchmarks
- 1.3.1 The basic batch forecast evaluation workflow in a nutshell - function metric interface
- 1.3.2 The basic batch forecast evaluation workflow in a nutshell - metric class interface
- 2. Forecasters in
sktime
- lookup, properties, main families - 3. Advanced composition patterns - pipelines, reduction, autoML, and more
- 4. Extension guide - implementing your own forecaster
- 5. Summary
- Useful resources
- Forecasting with sktime - appendix: forecasting, supervised regression, and pitfalls in confusing the two
- The pitfalls of mis-diagnosing forecasting as supervised regression
- Pitfall 1: over-optimism in performance evaluation, false confidence in “broken” forecasters
- Pitfall 2: obscure data manipulations, brittle boilerplate code to apply regressors
- Pitfall 3: Given a fitted regression algorithm, how can we generate forecasts?
- How does
sktime
help avoid the above pitfalls?
- The pitfalls of mis-diagnosing forecasting as supervised regression
- Probabilistic Forecasting with
sktime
- Overview of this notebook
- Quick Start - Probabilistic Forecasting with
sktime
- What is probabilistic forecasting?
- Probabilistic forecasting interfaces in
sktime
- Metrics for probabilistic forecasts and evaluation
- Advanced composition: pipelines, tuning, reduction, adding proba forecasts to any estimator
- Useful resources
- Credits
- Hierarchical, Global, and Panel Forecasting with
sktime
- Window Splitters in Sktime
- Preliminaries
- Data
- Visualizing temporal cross-validation window splitters
- A single train-test split using
temporal_train_test_split
- Single split using
SingleWindowSplitter
- Sliding windows using
SlidingWindowSplitter
- Sliding windows using
SlidingWindowSplitter
with an initial window - Expanding windows using
ExpandingWindowSplitter
- Multiple splits at specific cutoff values -
CutoffSplitter
- A single train-test split using
Classification#
- Overview of this notebook
- 2.1.1 preferred format 1 -
pd-multiindex
specification - 2.1.2 preferred format 2 -
numpy3D
specification - 2.2.3 Time Series Classification - deployment vignette
- 2.2.4 Time Series Classification - simple evaluation vignette
- 2.2.5 Time Series Regression - basic vignettes
- 5.2.6 Time Series Clustering - basic vignettes
- 2.4.1 Primer on
sktime
transformers for feature extraction - 2.4.2 Pipelines for time series panel tasks
- 2.4.3 Using transformers to deal with unequal length or missing values
- 2.4.4 Tuning and model selection
- 2.4.5 Advanced Composition cheat sheet - AutoML, bagging, ensembles
- Multi-variate time series classification using a simple CNN
- Load a dataset
- Train a deep neural network classifier
- Grid Search
- Channel Selection in Multivariate Time Series Classification
- 1 Initialise the Pipeline
- 2 Load and Fit the Training Data
- 3 Classify the Test Data
- 4 Identify channels
- 5 Standalone
- 6 Distance Matrix
- Dictionary based time series classification in sktime
- Early time series classification with sktime
- Interval based time series classification in sktime
Regression#
To come!
Clustering#
Annotation#
Transformation#
- The Canonical Time-series Characteristics (catch22) transform
- Feature extraction with tsfresh transformer
- Time series interpolating with sktime
- Now the interpolator enters
- MiniRocket
- Demo of the PlateauFinder transformer
- Demo of ROCKET transform
- The Signature Method with Sktime
- Theta Lines transformer
Data#
- In-memory data representations and data loading
- Section 1: in-memory data containers
- Section 1.1: Time series - the
"Series"
scitype - Section 1.1.1: Time series - the
"pd.DataFrame"
mtype - Section 1.1.2: Time series - the
"pd.Series"
mtype - Section 1.1.3: Time series - the
"np.ndarray"
mtype - Section 1.2: Time series panels - the
"Panel"
scitype - Section 1.2.1: Time series panels - the
"pd-multiindex"
mtype - Section 1.2.2: Time series panels - the
"numpy3D"
mtype - Section 1.2.3: Time series panels - the
"df-list"
mtype - Section 1.3: Hierarchical time series - the
"Hierarchical"
scitype
- Section 1.1: Time series - the
- Section 2: validity checking and mtype conversion
- Section 3: loading pre-defined data sets
- Section 4: loading data from
csv
files
- Section 1: in-memory data containers
- Loading data into sktime