h-iaac / har_flower

Toward a Federated Model for Human Context Recognition on Edge Devices

This Repository contains the code used in the homonymous paper to be published after the CBA (Congresso Brasileiro de Automática) conference.

Contents:

• module es_utils, containing classes and custom functions to build the model we used alongside the Extrasensory dataset
• folder code containing:
  • model folder with all models we used as base-model
  • server.py file with the code to run the server for Federated Leaning
  • client_har.py file with the code to run the client for Federated Leaning
  • exploratory_analysis.ipynb notebook with the an exploratory analysis of the dataset
  • experimento_1_base_model.ipynb notebook with the code to run the experiment run_federated.sh bash script to run the federated learning. Receives the number of the experiment fold and the path-to-folder to search the clients' data
  • unzip_all_csv.sh bash script to unzip all the csv files in the dataset
• folder sample_data containing the a small fraction of the data used in the experiment
• file requirements.txt with the list of packages used in the experiment

Requirements:

As seen in file requirements.txt, the following packages are required to run the code:

• numpy = "^1.23.0"
• pandas = "^1.4.3"
• sklearn = "^0.0"
• tensorflow = "^2.9.1"
• keras-tuner = "^1.1.2"
• flwr = "^0.17.0"

How to run:

After installing all dependencies and build the base models (by running the notebooks in the code folder), you can run the federated learning by running:

• the server by running the command: python server.py
• the bash script run_federated.sh. It receives two arguments:
  • the number of the experiment fold (0 to 4)
  • the path-to-folder to search the clients' data

Citation:

The original paper can be found in the pre-proceedings here under the name Toward a Federated Model for Human Context Recognition on Edge Devices. It still doesn't have a DOI, but it will have it soon.