liu-tianhang / dance-moves-determiner

💃 NUS CG3002 Embedded Systems Design Project (Software)

dance-moves-prediction

We aim to predict dance moves using a combination of Random Forest and Support Vector Machine (SVM) to get the best possible accuracy. Train and test data are based on the readings obtained from two sensors (GY-521 MPU6050 3-Axis Acceleration Gyroscope 6DOF Module) attached to the dancers wrists. The raw features are Acceleration x, y, z and Rotation x, y, z. There are a total of 11 dance moves to be classified.

Resolving Dependencies

1. Create a virtual environment in the directory

python3 -m venv venv

2. Activate the virtual environment

source venv/bin/activate

3. Install all the required dependencies

pip install -r requirements.txt

Generating Machine Learning Models

1. Activate the virtual environment
2. Resolve all the dependencies
3. Run python train_random_forest.py to generate Random Forest model
4. Find the models in .sav format in models directory
5. Find the training statistics and evaluation results in eval_results directory

To generate SVM model, run python train_svm.py at Step 3 instead.

Alternatively, run ./fast_generate_models.sh at Step 3 to generate both models at the same time.

See sample evaluation results here

Note that if you run this without GPU, each Machine Learning model takes around 1.5 hours to generate owing to the hyperparameter tuning involved. As such, you may want to train the models from a detached session of a terminal multiplexer.

Plotting Graphs for All Dance Moves

1. Activate the virtual environment
2. Resolve all the dependencies
3. Make sure that the code below is commented out in train_data.py 👎

  ##### COMMENT OUT THE CODE BELOW BEFORE RUNNING plot.py #####
  df_max_min = prep.flatten(df, interval)
  df_var = prep.flatten(df, 'var', interval)
  df_concat = prep.concat_df(df_max_min, df_var)
  df = prep.append_truth(df_concat, number)
  #############################################################

4. Run python plot.py
5. Find the graphs in .png format in plots directory

For some reason, adding an additional parameter in the dode to skip Step 3 makes the code runs 99999x slower. 😓

See sample plots here

Testing Final Model

1. Activate the virtual environment
2. Resolve all the dependencies
3. Generate machine learning models
4. Customise how the models generated are utilised in MLModel class (test_model/__init__.py file)
5. Run python test.py

Note that final model can be a combination of both Random Forest and SVM.