I'm trying to cluster an unknown set of data with a replicator neural network. The number of clusters is determined by the number of neuron units in the middle layer, multiplied by the number of steps in our step function responsible for the middle layer activation. The only information provided, is that the feature size vector will be 10. However there are no information about the number of possible clusters. The final goal is to detect anomalies with the neural network.
How do I approach this task? I'm not sure how much neurons and steps I should be using. I believe I can limit the neurons to about 4, as it should be usually less than half the input neurons. However I'm not sure how the network is going to perform on a lot of steps. Perhaps to many neurons won't hurt, as for my understand neural networks cluster, compared to k-means, independent from the number of clusters, while a number to small might damage the anomaly detection by limitation?