#minibatches

A validation set has $N = 1000$ examples, read in minibatches of 300, so the last minibatch holds only 100. The quick estimate $L^\text{q}$ averages the four minibatch mean losses with equal weight and equals 0.5; the last minibatch's mean loss is $l^\text{b} = 0.9$. What is the exact per-example validation loss $L$?