... | ... | @@ -48,7 +48,7 @@ How does the model work? As the name implies, we are defining cluster boundaries |
|
|
|
|
|
In k-means, we saw how we can apply an iterative algorithm to learn the stereotype cluster centres. In GM models, we use [Expectation-Maximization (EM) algorithm]( https://en.wikipedia.org/wiki/Expectation%E2%80%93maximization_algorithm). The idea behind is the same. In the expectation step, we again calculate the r matrix, but this time we find the probabilities to belong to a class for each instance, by using the assumed mean and variances (stereotypes). Note that we do not have one, but several Gaussians to represent the cluster distribution so we need to initialize a weight vector (mixing coefficients), which will combine the Gaussians according to their weights. Then for each cluster, we will find the new stereotype properties:
|
|
|
|
|
|
<img src="uploads/ea5baf619f11295c145bb1639e46a8d4/cluster_4.png" width="600">
|
|
|
<img src="uploads/ea5baf619f11295c145bb1639e46a8d4/cluster_4.png" width="400">
|
|
|
|
|
|
|
|
|
|
... | ... | |