#subsampling

When training word embeddings, each occurrence of a word $w_i$ is discarded with probability $\max\left(1 - \sqrt{t / f(w_i)},\ 0\right)$, where $f(w_i)$ is the word's share of all tokens and $t = 10^{-4}$. The word "the" makes up 4% of the corpus. What fraction of its occurrences is discarded? Round to 2 decimals.