model_train

Model Training

Usage

1. Push all the files that you edited to GitHub.
2. If your repo on GitHub is public,
   you can run the jupyter notebook in https://colab.research.google.com/github/<your_username>/<your_repo_name>/blob/master/src/model_train/train_magic_wand_model.ipynb
   where <your_username> may be NTHUEE240500 and <your_repo_name> may be mbed07.
   Otherwise, you may check Loading Private Notebooks section in Using Google Colab with GitHub.
3. Replace the you_repo_url and your_repo_name in the Google colab to satify the situation you have.
4. Press the play button beside the cell to run the code in the cell.

Note: You can tune some parameters in the config.py

DATA_NAME = "accel_ms2_xyz"
LABEL_NAME = "gesture"

# label name (you should keep "negative" in the end of the list)
labels = ["ring", "slope", "negative"]

# data split configuration
# note that train_ratio + valid_ratio + test_ratio = 1
train_ratio = 0.6
valid_ratio = 0.3
data_split_random_seed = 30

# model configuration
model = "CNN"
seq_length = 64 # the input size of the model
epochs = 50
steps_per_epoch =1000
batch_size = 64