In common_utils.py, there are some parameters you may want to set. It contains N, d, K (number of data points, dimension, number of clusters) and some hyperparameters. The code should be able to handle any value of d, but right now we are plotting stuff, and the code can only plot up to 2 dimensions. So if d > 2, you should turn off the plotting.
After having set your parameters, you can generate data using python data_generator.py. You will get print outs of parameter values and plots of the data you just generated. It saves the data to a file called data.txt. If you look inside data_generator.py, you can always figure out how to plot data from data.txt if you generated it a while back and forgot what it looks like.
To run the VI/SVI algorithms, simply run:
$:~ python variational_inference.py$:~ python stochastic_variational_inference.py.
You must make sure that data.txt is in the current directory. Look through the code for more details. We also include a comparison script to compare batch VI vs. minibatch SVI. You can run this by running:
$:~ python compare_VI_SVI.py
These files will give plots that were shown in the tutorial.