#split-selection

Training data in two features $(f_1, f_2)$: negatives $(-1, 0)$, $(2, 1)$, $(2, -2)$; positives $(0, 0)$, $(1, 0)$. Candidate tests are thresholds halfway between adjacent distinct values of one feature. The tree picks the test with the lowest average entropy $\sum_{\text{sides}} \frac{n_\text{side}}{n} H(\text{side})$ (base-2 logarithms). What is that lowest average entropy? Round to 3 decimals.