A*HAR: A NEW BENCHMARK TOWARDS SEMI-SUPERVISED LEARNING FOR CLASS IMBALANCED HUMAN ACTIVITY RECOGNITION

The tensorflow code for "A*HAR: A NEW BENCHMARK TOWARDS SEMI-SUPERVISED LEARNING FOR CLASS IMBALANCED HUMAN ACTIVITY RECOGNITION " (https://arxiv.org/abs/2101.04859)

The environment can be found in dockerhub: docker pull loklu/mt_tensorflow:tf1.2.1_py35_lib2

The code runs on Python 3. Install the dependencies and prepare the datasets with the following commands:

pip install tensorflow==1.2.1 numpy scipy pandas
./prepare_data.sh

To train the model, run:

• 'python /experiments/har/har.py --template X' for HAR models. modify arguments in template.py for choosing type of experiment and no: of unlabeled samples , Where X can be defined as suitable template