What's the difference between "relative" and "probability" for operator sample?

AP

What's the difference between "relative" and "probability" for operator sample?

The objective effect is the same

RolandJones

Hi @AP,

The relative options guarantees a certain proportion of the input dataset on output. For example, with 100 examples and a sample ratio of 0.5, I will get 50 examples on output. The probability option instead assigns a probability to each example of it being included in the sample, so you could get in theory and value between 0 and 100. In practice it will likely fall between 50 plus of minus a couple of standard deviations.

Hope this helps.

Best,

Roland