suppose i have 10 individual observations each of size (125,59). i want to group these 10 observations based on their 2d feature matrices ((125,59)).Is this possible without flattening every observation to 125*59 1D matrix ? I cant even implement PCA or LDA for feature extraction because the data is highly variant. Please note that i am trying to implement clustering through self organizing maps or neural networks. Deep learning and neural networks are completely related to the question asked.
Of course it is.
Define an appropriate distance measure.
Then compute the 10x10 distance matrix, and run hierarchical clustering.
来源：https://stackoverflow.com/questions/54815040/how-to-do-clustering-when-the-shape-of-data-is-x-y-z