Problem Definition

• A machine learning model that can recognize and classify different types of activities using smartphone sensor data

• Use different tools/frameworks to analyse data

Run the experiment

• setup:

  pip install --upgrade -r "requirements.txt"

• data preparation: (execute in the project root directory)

  # cleaning.py INPUT_DIR MIN_INTERVAL_SEC
  spark-submit --driver-memory 3G src/cleaning.py data/all_data_v2 60

Project Structure

• PROJ_ROOT:

  .
  ├── data/
  │   └── all_data_v2
  ├── model/
  │   ├── classifier.trs
  │   └── vae.trs
  ├── module/
  │   ├── __init__.py
  │   ├── nets.py
  │   └── util.py
  ├── src
  │   ├── activity_recognition.py
  │   ├── addLabel.py
  │   ├── cleaning.py
  │   ├── RFC_act_rec.py
  │   └── torch/
  │       ├── exp_torch.py
  │       ├── __init__.py
  │       ├── loader.py
  │       └── pre_loader.py
  ├── pyproject.toml
  ├── README.md
  └── requirements.txt
  

Running the SKLearn Models

The full dataset can be downloaded from the following link: https://drive.google.com/drive/folders/1qsQ0GcVMYLuoDPEXPlTfmGtm_HmPXZpI?usp=share_link

1. After cloning the repo and cd to the project root directory, you can start running the cleaning code.

   spark-submit src/cleaning.py example_data 60

   • Optional: add the argument –driver-memory [#]g replacing the [#] with amount of memory you choose to run the program with
     • With our full dataset we found 3g was the sweet spot

2. After running the cleaning script you should notice a folder called example_data-oput_60

   • This contains the grouped data that will train the sklearn models

3. Run the command

   python3 src/RFC_act_rec.py example_data-oput_60s example_ML_testing_data-oput_60s

   • This will create and test RandomForest and MLPClassifier models and output their results to src/output_pd/
   • The second argument in the example ('example_ML_testing_data-oput_60s') has to contain the cleaned data files used for testing, the files currently in 'example_ML_testing_data' are testing files for your convienience that don't overlap with the files provided in example_data/
     • These files can be generated manually by following the first step and replacing 'example_data' with 'example_ML_testing_data'

Running the PyTorch Models

1. After cloning the repo and cd to the project root directory, you can start running the cleaning code.

   spark-submit src/cleaning.py example_data 60 True

   • Optional: add the argument –driver-memory [#]g replacing the [#] with amount of memory you choose to run the program with
     • With our full dataset we found 3g was the sweet spot

2. After running the cleaning script you should notice a folder called example_data-oput_60

   • This contains the grouped data for use with the sklearn model

3. Run the command (in the project root directory)

   PYTHONPATH=./ python3 src/torch/exp_torch.py example_data-oput_60s

   • This will create and test RandomForest and MLPClassifier models and output their results to src/output_pd/