Implements approach from this paper and code from
here.
Can be used if labeled data contains noise, i.e. some of the data is labeled wrong.
Usage
noisy_loss_wrapper(noise_matrix)
Arguments
- noise_matrix
Matrix of noise distribution.
Value
A function implementing noisy loss.
Examples
if (FALSE) { # reticulate::py_module_available("tensorflow")
# If first label contains 5% wrong labels and second label no noise
noise_matrix <- matrix(c(0.95, 0.05, 0, 1), nrow = 2, byrow = TRUE)
noisy_loss <- noisy_loss_wrapper(noise_matrix)
}
