Replace context.yaml with logging information

[Deathn0t]

The quick start example in the documentation is shown below:

from deephyper.hpo import HpProblem, CBO
from deephyper.analysis.hpo import parameters_at_max
from deephyper.evaluator import Evaluator

def run(job):
    x = job.parameters["x"]
    b = job.parameters["b"]
    function = job.parameters["function"]

    if function == "linear":
        y = x + b
    elif function == "cubic":
        y = x**3 + b

    return y

def optimize():
    problem = HpProblem()
    problem.add_hyperparameter((-10.0, 10.0), "x")
    problem.add_hyperparameter((0, 10), "b")
    problem.add_hyperparameter(["linear", "cubic"], "function")

    evaluator = Evaluator.create(run, method="process",
        method_kwargs={
            "num_workers": 2,
        },
    )

    search = CBO(problem, evaluator, random_state=42)
    results = search.search(max_evals=100)

    return results

if __name__ == "__main__":
    results = optimize()
    print(results)

    res = parameters_at_max(results)
    print("\nOptimum values")
    print("function:", res[0]["function"])
    print("x:", res[0]["x"])
    print("b:", res[0]["b"])
    print("y:", res[1])

Running this example creates a context.yaml file. The purpose of this file is to summarize the optimization problem that was performed along with search and evaluator parameters among other things. This content should be logged to the screen or printed to a text file by the logger. There is no need to use YAML for this. Also, by replacing this functionality with logging features, it may be possible to remove the pyyaml dependency since it is not used elsewhere in the deephyper package.

[wigging]

In the Search class, there is a dump_context method that writes to the context.yaml file. Is there any code in deephyper that actually uses this context.yaml file or depends on its contents?

[Deathn0t]

The context.yaml is mostly used as a log. No other code within deephyper depends on it.

[wigging]

The context.yaml contains the following:

  problem: '{"name": null, "hyperparameters": [{"type": "uniform_int", "name": "b",
    "lower": 0, "upper": 10, "default_value": 5, "log": false, "meta": null}, {"type":
    "categorical", "name": "function", "choices": ["linear", "cubic"], "weights":
    null, "default_value": "linear", "meta": null}, {"type": "uniform_float", "name":
    "x", "lower": -10.0, "upper": 10.0, "default_value": 0.0, "log": false, "meta":
    null}], "conditions": [], "forbiddens": [], "python_module_version": "1.2.0",
    "format_version": 0.4}'

The key is problem and the value is a dictionary inside of a string. This appears to be caused by the get_init_params_as_json function in _search.py where JSON is used to dump an object's parameters or attributes to a string. Since this is a string it makes it harder to format and print the value to the screen. Can you tell me which objects are passed to this function so I can improve their representation for better screen printing?

[Deathn0t]

the serialization is for the problem argument of the __init__ of the search. This calls HpProblem.to_json() defined at this line. The self._space here is a ConfigSpace object that already has a nice __repr__ that you can try.

[wigging]

Is the context.yaml discussed anywhere in the documentation? Is there anything that tells users about the file and what it contains? I don't see anything in the documentation but maybe I missed it.