# Gaussian Mixture Classification Implementation with multidimensional trainning data

I'm trying to implement the gaussian mixture classification (GMC) implementation from scratch using python. The training dataset consists of 10 folds each of size $$\left[100x64\right]$$. In addition, each fold has it own train label of size $$\left[100x1\right]$$ with classes $$C=\left\{ 5,6\right\}$$. I'm stuck on the parameter estimations $$\mu_{i}$$, which according to the book "Bishop - Pattern Recognition And Machine Learning - Springer 2006", should be $$\mu_{1},\mu_{2}$$ but I have been told that both should be a vector of $$\left[1x64\right]$$ I first need to estimate optimal parameters before maximizing likelihood but I don't know-how.