Must a dataset for a classifier be perfectly balanced?

Question

In machine learning it is generally accepted that the dataset for a binary classifier has to be balanced: EG. class 1 50%, class 0 50%. My question is: has the dataset to be