BigDL-Chronos (Chronos in short) is an application framework for building a fast, accurate and scalable time series analysis application, where forecasting is the most popular task. For more detailed information about the framework, go to Chronos Document.
-
Chronos provides an easy but complete API
TSDataset
for data loading, preprocessing, feature engineering and sampling, where users could fetch their raw tabular data from local files, distributed file system or TS database like Prometheus and transform them to data that can be fed into DL/ML models. It contains 2 backends, pandas for single node users and spark (with the support ofbigdl-orca
) for cluster users. -
Chronos provides
Forecaster
API for 9+ different ML/DL models' training, tunning, evaluation, model optimization, and exporting for deployment. -
9+ different ML/DL model include traditional models cover traditional models like ARIMA, Prophet to SOTA Deep learning models such as Autoformer, TCN, NBeats, and Seq2Seq.
-
During the process, IPEX, intel-tensorflow, onnxruntime, openvino, and intel neural compressor are used as accelerator and low precision tool to accelerate the model during training and inferencing.
bigdl-orca
is used for cluster level distributed training and tunning.
bigdl-chronos
is released on PyPI as a python library and could be installed on nearly any platform user prefer. Here we provide a typical installation method for single node user who prefer a pytorch backend, for more detailed information, users may refer to our installaion page.
# conda environment is highly recommeneded
conda create -n bigdl-chronos-pytorch-env python=3.7 setuptools=58.0.4
conda activate bigdl-chronos-pytorch-env
pip install bigdl-chronos[pytorch]
Users could also install bigdl-chronos
easily on google colab, where they only need to install the library through pip
.
Some users may prefer a docker installation do seperate the environment. Below setup and how-to-run sessions are for users who want to use the provided docker image.
Pull Docker Image
docker pull intel/ai-workflows:time-series-forecaster
To run the pipeline, follow the instructions below outside of docker instance.
docker run -it --rm --net=host intel/ai-workflows:time-series-forecaster \
/bin/bash jupyter nbconvert --to python \ chronos_nyc_taxi_tsdataset_forecaster.ipynb && \
sed '26,40d' chronos_nyc_taxi_tsdataset_forecaster.py > chronos_taxi_forecaster.py && \
python chronos_taxi_forecaster.py
How-to guides are bite-sized, executable examples where users could check when meeting with some specific topic during the usage. For friendly experience please visit our how to guide page.
Here is a use-cases list of bigdl-chronos
, for more friendly experience please visit our tutorial page.
-
Intel® Xeon® Scalable Performance processors
-
Intel® Core® processors
-
Ubuntu 16.04/18.04/20.04/22.04
-
Windows (experimentally supported)
-
Mac with Intel Chip (experimentally supported)
-
Python 3.7
-
Python 3.8 (experimentally supported)
We welcome any questions, bug report or feature request to: