One Change at a Time

You're trying out settings for a model. The starter code has train(depth, seed, rate=0.1), which just describes the run as text.

Task: write run_all(base, changes).

• base is a dictionary of starting settings, like {"depth": 3, "seed": 0}.
• changes is a list of dictionaries. Each one says which settings to change for one run.
• For each change, build that run's settings: everything in base, with the change's values on top. Then call train with those settings by name.
• Return the list of what train gave back, one per change, in order.
• Don't change base itself — each run starts again from base.