Tensorflow MixtureSameFamily and gaussian mixture model

Question

I am really new to Tensorflow as well as gaussian mixture model.

I have recently used tensorflow.contrib.distribution.MixtureSameFamily class for predicting probability density function which is derived from gaussian mixture of 4 components.

When I plotted the predicted density function using "prob()" function as Tensorflow tutorial explains, I found the plotted pdf with only one mode. I expected to see 4 modes as the mixture components are 4.

I would like to ask whether Tensorflow uses any global mode predicting algorithm in their MixtureSameFamily class. If not, I would also like to know how MixtureSameFamily class forms the pdf with statistical values.

Thank you very much.

Answer 1

I found an answer for above question thanks to my collegue.

The 4 components of gaussian mixture have had very similar means that the mixture seems like it has only one mode.

If I put four explicitly different values as means to the MixtureSameFamily class, I could get a plot of gaussian mixture with 4 different modes.

Thank you very much for reading this.