Hierarchical clustering does not work for too many points (>1e5).
We need to take a sample of N power load points per appliance (N=1e4).
It is necessary to find a method that ensures the sample of N points contains at least one activation of each potential status. Otherwise, we risk taking a sample of points full of 0's (because many appliances are inactive >90% of the recording period).
Hierarchical clustering does not work for too many points (>1e5).
We need to take a sample of N power load points per appliance (N=1e4).
It is necessary to find a method that ensures the sample of N points contains at least one activation of each potential status. Otherwise, we risk taking a sample of points full of 0's (because many appliances are inactive >90% of the recording period).