Average imputation accuracies of the seven methods, fastPHASE (fPH), NN, WNN, SVM, MC, NeuralNet (NeuN) and BaseLine (BL), on the two real cattle datasets. fastPHASE performed statistically significantly better (p-values < 0.0001) than all the other six methods on both datasets; our methods NN and WNN and SVM performed no differently, and significantly better (p-values < 0.0001) than MC, NeuralNet and BaseLine. Interestingly, it is challenging to argue which one of the two datasets is easier to impute. As far as we know, the low density dataset has been curated multiple times when it was used in GWAS, and it contains much more samples (469) which are certainly helpful for imputation; the high density dataset was more recently generated, has not been used in GWAS, and it contains only 64 samples. Nevertheless, the performance differences between fastPHASE and the second group of NN, WNN and SVM, and between the second group and the other three methods are similar to those on high density human datasets.