Installation

Clone the repository from GitHub:

git clone https://github.com/naka-lab/Serket.git

Install dependency packages (you can skip this step if you have already installed these packages):

pip install numpy
pip install opencv-python
pip install numba
pip install scipy
pip install tensorflow
pip install librosa==0.5.1

Simple Example of Serket

First, the toy dataset of six observations is generated. Here, these are assumed to be generated from three categories of latent variables.

data = [
    [10, 2, 1],
    [8, 1, 1],
    [2, 7, 0],
    [1, 11, 2],
    [1, 1, 8],
    [2, 1, 1]
]

data_category = [0, 0, 1, 1, 2, 2]

Next, we define the modules. srk.Observation is a module that sends observations to another module and mlda.MLDA (multimodal latent Dirichlet allocation) is a module for unsupervised classification. Here, we define a kind of MLDA that classifies the dataset into three classes. By using the optional argument category, the classification accuracy can be computed automatically.

obs = srk.Observation( data )
mlda1 = mlda.MLDA( 3 , category=data_category )

By connecting modules, the model is constructed and the parameters of the model are estimated by the update method.

mlda1.connect( obs )
mlda1.update()

If the model was successfully trained, you can find the module000_mlda directory in your working directory, which contains the training results. The classification accuracy that is computed automatically is saved in module000_mlda/000/acc.txt. It might indicate a value of 1.00, which represents the all observations were classified correctly.