As far as I know, there is no such rules of thumb to determine the ratio between training sample and testing sample. Intuitively, more training data should not hurt because the selected model may approximate the real world data well if my data are a good representative of the population.
↧