1. Root Mean square Error RMSE (Root Mean Squared Error)

2. Root-mean-square logarithm Error RMSLE (Root Mean Squared MIC Error)

Advantages of using RMSLE
1.RMSLE penalty underprediction is larger than overprediction, which is applicable to some scenarios where underprediction is more costly, such as the prediction of bike-sharing demand.
If the true value is 1000, if the predicted value is 600, then RMSE=400, RMSLE=0.510
if the true value is 1000, if the predicted value is 1400, then RMSE=400, RMSLE=0.336
It can be seen that under the same RMSE condition, RMSLE with a predicted value smaller than the true value is larger, that is, the penalty for a predicted value smaller is larger.
2. If the range of predicted values is large, RMSE will be dominated by some large values. So even if you get a lot of small value predictions right, but one very large value prediction wrong, RMSE will be very large. Accordingly, if another poor algorithm is more accurate for this large value, but many small values are biased, it is possible that RMSE will be smaller than the previous one. You can solve this problem a little bit by taking log first and then RMSE. RMSE is generally a reasonable predictor of a fixed mean distribution.