Dependencies: Python version: 3.6+ (for used packages see utils.py) Remark: This code is largely based on the implementations from the Flow Data ML Repositories.
The Flow Data Streaming Client Ensemble realizes the Normalization step of the Data Preparation stage as well as the Machine Learning stage (see Flow Data Streaming Pipeline). Prepared flow data is received from the Flow Data Streaming Server and further preprocessed (feature filtering, feature selection, normalization, labeling) and used for online ensemble testing and training. In general, any online model can be implemented and used at the client side.
The Flow Data Streaming Client Ensemble can be started as follows:
- test and train an online ensemble of fully-connected DNNs on streamed, prepared flow data:
python3 ensemble-client.py
While parameters for the server connection handler are specified in constants.py (e.g., address information), all experiments using the ensemble client can be configured by command line parameters (see defaultParser.py).
Prediction of Recurrent Flow Characteristics Using Time Scale Ensemble Learners, PNM 2021