Handle configspace as dictionary in mlp example

[sarah-segel]

• Handle configspace as dictionary in mlp and parego example
• Adapt sgd loss to newest scikit-learn version ("log" not accepted anymore")

[eddiebergman]

You should look at the definition of Mapping.get which applies to dicst and to Configuration now. It's roughly something like:

def get(self, key: str, default = None):
    try:
        return self[key]
    except KeyError:
        return default

From an Ipython session

In [1]: d = {}

In [2]: d.get?
Signature: d.get(key, default=None, /)
Docstring: Return the value for key if key is in the dictionary, else default.
Type:      builtin_function_or_method

[eddiebergman]

Just an FYI, you can do

lr = config.get("learning_rate", "constant")
lr_init = config.get("learning_rate_init", 0.001)
batch_size = config.get("batch_size", 200)

[eddiebergman]

If n_neurons is going to be in there, you can still do config["n_neurons"]. In fact it would error if either of these were to not be present as [None] * None won't work. Likewise for the two lines below this one.

You use get(...) when there's a sensible default to use, otherwise you should do config[key] when there is no sensible default and it should error if it's not present.