Example for custom parameter passing in surrogate models¶

This example shows how to define surrogate models with custom model parameters. It also shows the validations that are done and how to specify these parameters through a configuration.

This example assumes some basic familiarity with using BayBE. We thus refer to campaign for a basic example.

Necessary imports¶

import numpy as np

from baybe.campaign import Campaign
from baybe.objective import Objective
from baybe.parameters import (
    CategoricalParameter,
    NumericalDiscreteParameter,
    SubstanceParameter,
)
from baybe.recommenders import (
    FPSRecommender,
    SequentialGreedyRecommender,
    TwoPhaseMetaRecommender,
)
from baybe.searchspace import SearchSpace
from baybe.surrogates import NGBoostSurrogate
from baybe.targets import NumericalTarget
from baybe.utils.dataframe import add_fake_results

Experiment Setup¶

parameters = [
    CategoricalParameter(
        name="Granularity",
        values=["coarse", "medium", "fine"],
        encoding="OHE",
    ),
    NumericalDiscreteParameter(
        name="Pressure[bar]",
        values=[1, 5, 10],
        tolerance=0.2,
    ),
    NumericalDiscreteParameter(
        name="Temperature[degree_C]",
        values=np.linspace(100, 200, 10),
    ),
    SubstanceParameter(
        name="Solvent",
        data={
            "Solvent A": "COC",
            "Solvent B": "CCC",
            "Solvent C": "O",
            "Solvent D": "CS(=O)C",
        },
        encoding="MORDRED",
    ),
]

Create a surrogate model with custom model parameters¶

Please note that model_params is an optional argument: The defaults will be used if none specified

surrogate_model = NGBoostSurrogate(model_params={"n_estimators": 50, "verbose": True})

Validation of model parameters¶

try:
    invalid_surrogate_model = NGBoostSurrogate(model_params={"NOT_A_PARAM": None})
except ValueError as e:
    print("The validator will give an error here:")
    print(e)

The validator will give an error here:
Invalid model params for NGBoostSurrogate: NOT_A_PARAM.

Links for documentation¶

Note that GaussianProcessSurrogate will support custom parameters in the future

RandomForestModel NGBoostModel BayesianLinearModel

Creating the campaign¶

campaign = Campaign(
    searchspace=SearchSpace.from_product(parameters=parameters, constraints=None),
    objective=Objective(
        mode="SINGLE", targets=[NumericalTarget(name="Yield", mode="MAX")]
    ),
    recommender=TwoPhaseMetaRecommender(
        recommender=SequentialGreedyRecommender(surrogate_model=surrogate_model),
        initial_recommender=FPSRecommender(),
    ),
)

Iterate with recommendations and measurements¶

We can print the surrogate model object

print("The model object in json format:")
print(surrogate_model.to_json(), end="\n" * 3)

The model object in json format:
{"type": "NGBoostSurrogate", "model_params": {"n_estimators": 50, "verbose": true}}

# Let's do a first round of recommendation
recommendation = campaign.recommend(batch_size=2)

print("Recommendation from campaign:")
print(recommendation)

Recommendation from campaign:
    Granularity  Pressure[bar]  Temperature[degree_C]    Solvent
2        coarse            1.0                  100.0  Solvent C
239      medium           10.0                  200.0  Solvent D

# Add some fake results
add_fake_results(recommendation, campaign)
campaign.add_measurements(recommendation)

Model Outputs¶

Note that this model is only triggered when there is data.

print("Here you will see some model outputs as we set verbose to True")

Here you will see some model outputs as we set verbose to True

# Do another round of recommendations
recommendation = campaign.recommend(batch_size=2)

[iter 0] loss=1.0724 val_loss=0.0000 scale=1.0000 norm=0.7071

Print second round of recommendations

print("Recommendation from campaign:")
print(recommendation)

Recommendation from campaign:
      Granularity  Pressure[bar]  Temperature[degree_C]    Solvent
index                                                             
6          coarse            1.0             111.111111  Solvent C
10         coarse            1.0             122.222222  Solvent C

Using configuration instead¶

Note that this can be placed inside an overall campaign config Refer to create_from_config for an example

Note that the following explicit call str() is not strictly necessary. It is included since our method of converting this example to a markdown file does not interpret this part of the code as python code if we do not include this call.

CONFIG = str(
    """
{
    "type": "NGBoostSurrogate",
    "model_params": {
        "n_estimators": 50,
        "verbose": true
    }
}
"""
)

### Model creation from json
recreate_model = NGBoostSurrogate.from_json(CONFIG)

This configuration creates the same model

assert recreate_model == surrogate_model