The warm-up and forecast sample is still unclear to me. Do you compute the MSE, MAD, MAPE, of the warm up sample or only compute such using the forecast sample? What I did was separate the data of warm up and forecast (as in separate tables), then compute MSE, MAD, MAPE of the two sample.

So for example there are 31 periods of data, 16 periods is the warm up while 15 periods is the forecast. Do I perform the forecast with the whole dataset and just get the forest accuracy measures of the forecast sample? Then what was the significance of the warm up sample??? I'm hoping to find an illustration to explain this concept clearer to me such as in the case of naive and moving averages.