Predicting the firing rate of a cell with gradient boosted trees.

<p>Each row corresponds to the learning of one tree by the algorithm. The tuning curve is sequentially split as shown on the left figures (vertical lines; blue line displays the actual tuning curve and black lines correspond to the prediction). Thus, intervals between each pair of splits are assigned a different target value. The first two trees are shown on the right and the exact values of each leaf are indicated in the square boxes. Note that the predicted firing rates are the sum over all the leaves (i.e. the value of a single leaf can not be directly interpreted.