Okay, from the initial description, looks like MML looks at TOTAL length, where the message includes both the theory and the additional info needed to reconstruct the total data, while MDL ignores aspects of the description of the theory for the purposes of measuring the length.
I’m a bit confused on that point myself. Before finding that document, my understanding was that MML averaged over the prior, while MDL avoided having a prior by using some kind of minimax approach, but the paper I pointed you to doesn’t seem to say anything about that.
Try this.
Reading it now, thanks.
Okay, from the initial description, looks like MML looks at TOTAL length, where the message includes both the theory and the additional info needed to reconstruct the total data, while MDL ignores aspects of the description of the theory for the purposes of measuring the length.
Did I get that right or am I misunderstanding?
I’m a bit confused on that point myself. Before finding that document, my understanding was that MML averaged over the prior, while MDL avoided having a prior by using some kind of minimax approach, but the paper I pointed you to doesn’t seem to say anything about that.