Random Forest: majority vote aggregation

The problem, from first principles

One tree can make a brittle decision from its particular training data. A forest lets independently built trees vote, so patterns they agree on survive while individual tree mistakes can be outvoted. This question combines already-built classification trees; it does not build or traverse them itself.

From theory to code

Implement random_forest_predict(trees, input) by collecting each tree's predictions and taking a separate majority vote for every input row.

Constraints

• trees is a list of compatible fitted tree dictionaries and input has shape (n_samples, n_features).
• Return integer labels with shape (n_samples,).
• Reuse predict_tree once per tree; do not reimplement traversal.
• Vote independently down each prediction column.
• Break ties by the lower class label.

Hints

Open one at a time. Each gives away a little more than the last.