Meta-HAR

Meta-HAR: Federated Representation Learning for Human Activity Recognition
Paper published in TheWebConf 2021

Dataset

Details in ./data_process/readme.md

1. For collected dataset:

   cd data_process
   python feature_extraction.py --in_dir 'dir stores the original txt data' --out_dir 'dir which is used to store the pickle data'   

   The feature_extraction.py generates pickle files and the trans_dict_collect.pickle file.

2. For processing of the HHAR dataset please refer to: https://github.com/yscacaca/HHAR-Data-Process. To run on public dataset for yourself, make the dataset to have the same format as mentioned in the ./data_process/readme.md

Run

1. data process as mentioned above.
2. Run Meta-HAR with default hyper-parameters.

    python Central.py   # for central model. 
    python meta-har.py  # for meta-har

Note: Configure your own data and output dirs

Others

To run other baselines:

1. Reptile: Change the norm_embed to norm_cce in the Meta-HAR and remove fine-tune.
2. Meta-HAR-CE: Use "target" instead of "target_t" in fine-tune.

TODO

Processing code for the HHAR and the USC-HAD datasets.

Contact

Chenglin Li - [email protected]