Symmetric mean absolute percentage error

Source: R/metrics.R

Symmetric mean absolute percentage error

nnf_smape(input, target)

Arguments

input

(torch_tensor) A tensor of actual values

target

(torch_tensor) A tensor with the same shape as the input

Details

Computed according to the formula: $$SMAPE = \frac{1}{n}\displaystyle\sum_{t=1}^{n} \frac{\left\|input - target\right\|}{(\left\|target\right\| + \left\|input\right\|) *0.5}$$

See also

yardstick::smape

Examples

input <- c(92, 6.5, 57.69, 15.9, 88.47, 75.01, 5.06, 45.95, 27.8, 70.96)
input <- as_tensor(input)

target <- c(91.54, 5.87, 58.85, 10.73, 81.47, 75.39, 2.05, 40.95, 27.34, 66.61)
target <- as_tensor(target)

nnf_smape(input, target)
#> torch_tensor
#> 0.164419
#> [ CPUFloatType{} ]