Model selection as a function of the number of samples.

First row shows the size of the data set, 1000…28966, and the rest of the table shows the logarithm of the marginal likelihood of each model in the family for these data. Error bars on the log-marginal likelihoods are in S1 Table. As the number of samples increases, more complex models are required to explain the details of PC spiking, but the complexity eventually saturates, presumably having matched the complexity of the real cells observed at the given experimental accuracy.