auto-sktime

Automatic creation of time series forecasts, regression and classification.

Installation

For trouble shooting and detailed installation instructions, see the documentation.

Operating system: Linux
Python version: Python 3.8, 3.9, 3.10 and 3.11 (only 64 bit)
Package managers: pip

pip

auto-sktime is available in pip. You can see all available wheels here.

pip install auto-sktime

or, with maximum dependencies,

pip install auto-sktime[all_extras]

auto-sktime: Automated Time Series Forecasting

This section describes how to reproduce the results in the auto-sktime paper. First, install auto-sktime either via pip or from source as described above.

Next, switch to the scripts/benchmark directory and use

python benchmark.py

to benchmark all available methods on all datasets. Alternatively, you can also only execute the benchmark for selected methods and/or datasets. For example, pmdarima can be benchmarked on the very first dataset using

python benchmark.py --method pmdarima --end-index 1

Check

python benchmark.py --help

for detailed information how to configure the benchmark.

Reproducing results

To ensure fair comparisons due to parallel computing, the benchmark can and should be limited a single CPU core using

taskset --cpu-list 0 python benchmark.py

The raw results are going to be stored in CSV files on disk. To create the visualizations, use

python evaluation.py

after finishing the complete benchmark.

Remaining Useful Life Predictions (AutoRUL)

This section describes how to reproduce the results in the AutoRUL paper. First, checkout the exact code that was used to create the results. Therefore, you can use the tag v0.1.0

git checkout tags/v0.1.0 -b autorul

Next, switch to the scripts directory and use

python remaining_useful_lifetime.py <BENCHMARK>

to run a single benchmark data set. To view the available benchmarks and all configuration parameters run

python remaining_useful_lifetime.py --help

Reproducing results

You can use the following commands to recreate the reported baseline results in the experiments of the paper.

python remaining_useful_lifetime.py <BENCHMARK> --runcount_limit 1 --timeout 3600 --multi_fidelity False --include baseline_lstm
python remaining_useful_lifetime.py <BENCHMARK> --runcount_limit 1 --timeout 3600 --multi_fidelity False --include baseline_cnn
python remaining_useful_lifetime.py <BENCHMARK> --runcount_limit 1 --timeout 3600 --multi_fidelity False --include baseline_transformer
python remaining_useful_lifetime.py <BENCHMARK> --runcount_limit 1 --timeout 7200 --multi_fidelity False --include baseline_rf
python remaining_useful_lifetime.py <BENCHMARK> --runcount_limit 200 --timeout 7200 --multi_fidelity False --ensemble_size 1 --include baseline_svm

with <BENCHMARK> being one of {cmapss,cmapss_1,cmapss_2,cmapss_3,cmapss_4,femto_bearing,filtration,phm08,phme20}. For the AutoRUL evaluation only the benchmark is provided and all remaining default configurations are used.

python remaining_useful_lifetime.py <BENCHMARK>

To reproduce the results from AutoCoevoRUL, checkout the repository from Github and use the autocoevorul.py file to either export the data sets or import the results.

Note

This project has been set up using PyScaffold 4.2.1. For details and usage information on PyScaffold see https://pyscaffold.org/.

Building

To create a new release of auto-sktime you will have to install build and twine

pip install build twine
python -m build