hyper-parameters tuning

[MarcBresson]

hyper-parameters tuning:

• data augmentation parameters
• one skeleton vs two skeletons
• including limb angles or not
• window size
• classifier type (LSTM, FC neural network, knn, logistic regression, adaboost, gradientboost, randomforest)
• classifier hyper-parameters

see how to evaluate models. What metrics ? What should we maximize ?

[MarcBresson]

processed_data_augment_1
-> ut_interaction
   -> 10fps
       -> 0_11_4
          -> 1.json
          -> 2.json
          -> 3.json

Once data augmentation csv is generated, create features with python src/preprocessing_features.py

1. window size (it is used by the computation of velocity)
2. angle or not

After the feature file is generated, run the appropriate classifier

3. Classifier type
4. Classifier hyperparameters

Save history to evaluate the best model afterall