Unique solution for probabilistic PCA

[ferrine]

@dycontri

data_shape=obs.shape batch_size=64 if batch_size: obs=pm.Minibatch(obs, batch_size) else: batch_size=data_shape[0]

What happens if you increase number of samples to estimate the lower bound? obj_n_mc parameter, e.g. 5, 10, 100

or increase batch size?