Example for using a mixture use case in a discrete searchspace¶
Example for imposing sum constraints for discrete parameters. The constraints simulate a situation where we want to mix up to three solvents. However, their respective fractions need to sum up to 100. Also, the solvents should never be chosen twice, which requires various other constraints.
This example assumes some basic familiarity with using BayBE.
We thus refer to campaign for a basic example.
Necessary imports for this example¶
import math
import os
import numpy as np
from baybe import Campaign
from baybe.constraints import (
DiscreteDependenciesConstraint,
DiscreteNoLabelDuplicatesConstraint,
DiscretePermutationInvarianceConstraint,
DiscreteSumConstraint,
ThresholdCondition,
)
from baybe.objective import Objective
from baybe.parameters import NumericalDiscreteParameter, SubstanceParameter
from baybe.searchspace import SearchSpace
from baybe.targets import NumericalTarget
from baybe.utils.dataframe import add_fake_results
Experiment setup¶
This parameter denotes the tolerance with regard to the calculation of the sum.
SUM_TOLERANCE = 1.0
SMOKE_TEST = "SMOKE_TEST" in os.environ
# This parameter denotes the resolution of the discretization of the parameters
RESOLUTION = 5 if SMOKE_TEST else 12
dict_solvents = {
"water": "O",
"C1": "C",
"C2": "CC",
"C3": "CCC",
}
solvent1 = SubstanceParameter(name="Solv1", data=dict_solvents, encoding="MORDRED")
solvent2 = SubstanceParameter(name="Solv2", data=dict_solvents, encoding="MORDRED")
solvent3 = SubstanceParameter(name="Solv3", data=dict_solvents, encoding="MORDRED")
Parameters for representing the fraction.
fraction1 = NumericalDiscreteParameter(
name="Frac1", values=list(np.linspace(0, 100, RESOLUTION)), tolerance=0.2
)
fraction2 = NumericalDiscreteParameter(
name="Frac2", values=list(np.linspace(0, 100, RESOLUTION)), tolerance=0.2
)
fraction3 = NumericalDiscreteParameter(
name="Frac3", values=list(np.linspace(0, 100, RESOLUTION)), tolerance=0.2
)
parameters = [solvent1, solvent2, solvent3, fraction1, fraction2, fraction3]
Creating the constraint¶
Since the constraints are required for the creation of the searchspace, we create
them next.
Note that we need a PermutationInvarianceConstraint here.
The reason is that constraints are normally applied in a specific order.
However, the fractions should be invariant under permutations.
We thus require an explicit constraint for this.
perm_inv_constraint = DiscretePermutationInvarianceConstraint(
parameters=["Solv1", "Solv2", "Solv3"],
dependencies=DiscreteDependenciesConstraint(
parameters=["Frac1", "Frac2", "Frac3"],
conditions=[
ThresholdCondition(threshold=0.0, operator=">"),
ThresholdCondition(threshold=0.0, operator=">"),
ThresholdCondition(threshold=0.0, operator=">"),
],
affected_parameters=[["Solv1"], ["Solv2"], ["Solv3"]],
),
)
This is now the actual sum constraint
sum_constraint = DiscreteSumConstraint(
parameters=["Frac1", "Frac2", "Frac3"],
condition=ThresholdCondition(threshold=100, operator="=", tolerance=SUM_TOLERANCE),
)
The permutation invariance might create duplicate labels. We thus include a constraint to remove them.
no_duplicates_constraint = DiscreteNoLabelDuplicatesConstraint(
parameters=["Solv1", "Solv2", "Solv3"]
)
constraints = [perm_inv_constraint, sum_constraint, no_duplicates_constraint]
Creating the searchspace and the objective¶
searchspace = SearchSpace.from_product(parameters=parameters, constraints=constraints)
________________________________________________________________________________
[Memory] Calling baybe.utils.chemistry._smiles_to_mordred_features...
_smiles_to_mordred_features('C')
_______________________________________smiles_to_mordred_features - 0.2s, 0.0min
________________________________________________________________________________
[Memory] Calling baybe.utils.chemistry._smiles_to_mordred_features...
_smiles_to_mordred_features('CC')
_______________________________________smiles_to_mordred_features - 0.0s, 0.0min
objective = Objective(
mode="SINGLE", targets=[NumericalTarget(name="Target_1", mode="MAX")]
)
Creating and printing the campaign¶
campaign = Campaign(searchspace=searchspace, objective=objective)
print(campaign)
[1mCampaign[0m
[1mMeta Data[0m
Batches Done: 0
Fits Done: 0
[1mSearch Space[0m
[1mSearch Space Type: [0mDISCRETE
[1mDiscrete Search Space[0m
[1mDiscrete Parameters[0m
Name Type Num_Values Encoding
0 Solv1 SubstanceParameter 4 SubstanceEncoding.MORDRED
1 Solv2 SubstanceParameter 4 SubstanceEncoding.MORDRED
2 Solv3 SubstanceParameter 4 SubstanceEncoding.MORDRED
3 Frac1 NumericalDiscreteParameter 5 None
4 Frac2 NumericalDiscreteParameter 5 None
5 Frac3 NumericalDiscreteParameter 5 None
[1mExperimental Representation[0m
Solv1 Solv2 ... Frac2 Frac3
0 water C1 ... 0.0 100.0
1 water C1 ... 25.0 75.0
2 water C1 ... 50.0 50.0
.. ... ... ... ... ...
31 C1 C2 ... 25.0 50.0
32 C1 C2 ... 50.0 25.0
33 C1 C2 ... 25.0 25.0
[34 rows x 6 columns]
[1mMetadata:[0m
was_recommended: 0/34
was_measured: 0/34
dont_recommend: 0/34
[1mConstraints[0m
Type Affected_Parameters
0 DiscretePermutationInvarianceConstraint [Solv1, Solv2, Solv3]
1 DiscreteSumConstraint [Frac1, Frac2, Frac3]
2 DiscreteNoLabelDuplicatesConstraint [Solv1, Solv2, Solv3]
[1mComputational Representation[0m
Solv1_MORDRED_nHetero Solv1_MORDRED_AATSC1v ... Frac2 Frac3
0 1.0 -18.543836 ... 0.0 100.0
1 1.0 -18.543836 ... 25.0 75.0
2 1.0 -18.543836 ... 50.0 50.0
.. ... ... ... ... ...
31 0.0 -36.020386 ... 25.0 50.0
32 0.0 -36.020386 ... 50.0 25.0
33 0.0 -36.020386 ... 25.0 25.0
[34 rows x 12 columns]
[1mObjective[0m
[1mMode: [0mSINGLE
[1mTargets [0m
Type Name Mode Lower_Bound Upper_Bound Transformation \
0 NumericalTarget Target_1 MAX -inf inf None
Weight
0 100.0
[1mCombine Function: [0mGEOM_MEAN
TwoPhaseMetaRecommender(allow_repeated_recommendations=None,
allow_recommending_already_measured=None, initial_recommender=RandomRecommender(allow_repeated_recommendations=False, allow_recommending_already_measured=True), recommender=SequentialGreedyRecommender(allow_repeated_recommendations=False, allow_recommending_already_measured=True, surrogate_model=GaussianProcessSurrogate(model_params={}, _model=None), acquisition_function_cls=’qEI’, _acquisition_function=None, hybrid_sampler=’None’, sampling_percentage=1.0), switch_after=1)
Manual verification of the constraint¶
The following loop performs some recommendations and manually verifies the given constraints.
N_ITERATIONS = 2 if SMOKE_TEST else 3
for kIter in range(N_ITERATIONS):
print(f"\n#### ITERATION {kIter+1} ####")
print("## ASSERTS ##")
print(
"No. of searchspace entries where fractions do not sum to 100.0: ",
campaign.searchspace.discrete.exp_rep[["Frac1", "Frac2", "Frac3"]]
.sum(axis=1)
.apply(lambda x: x - 100.0)
.abs()
.gt(SUM_TOLERANCE)
.sum(),
)
print(
"No. of searchspace entries that have duplicate solvent labels: ",
campaign.searchspace.discrete.exp_rep[["Solv1", "Solv2", "Solv3"]]
.nunique(axis=1)
.ne(3)
.sum(),
)
print(
"No. of searchspace entries with permutation-invariant combinations: ",
campaign.searchspace.discrete.exp_rep[["Solv1", "Solv2", "Solv3"]]
.apply(frozenset, axis=1)
.to_frame()
.join(campaign.searchspace.discrete.exp_rep[["Frac1", "Frac2", "Frac3"]])
.duplicated()
.sum(),
)
# The following asserts only work if the tolerance for the threshold condition in
# the constraint are not 0. Otherwise, the sum/prod constraints will remove more
# points than intended due to numeric rounding
print(
f"No. of unique 1-solvent entries (exp. {math.comb(len(dict_solvents), 1)*1})",
(campaign.searchspace.discrete.exp_rep[["Frac1", "Frac2", "Frac3"]] == 0.0)
.sum(axis=1)
.eq(2)
.sum(),
)
print(
f"No. of unique 2-solvent entries (exp."
f" {math.comb(len(dict_solvents), 2)*(RESOLUTION-2)})",
(campaign.searchspace.discrete.exp_rep[["Frac1", "Frac2", "Frac3"]] == 0.0)
.sum(axis=1)
.eq(1)
.sum(),
)
print(
f"No. of unique 3-solvent entries (exp."
f" {math.comb(len(dict_solvents), 3)*((RESOLUTION-3)*(RESOLUTION-2))//2})",
(campaign.searchspace.discrete.exp_rep[["Frac1", "Frac2", "Frac3"]] == 0.0)
.sum(axis=1)
.eq(0)
.sum(),
)
rec = campaign.recommend(batch_size=5)
add_fake_results(rec, campaign)
campaign.add_measurements(rec)
#### ITERATION 1 ####
## ASSERTS ##
No. of searchspace entries where fractions do not sum to 100.0: 0
No. of searchspace entries that have duplicate solvent labels: 0
No. of searchspace entries with permutation-invariant combinations: 0
No. of unique 1-solvent entries (exp. 4) 4
No. of unique 2-solvent entries (exp. 18) 18
No. of unique 3-solvent entries (exp. 12) 12
#### ITERATION 2 ####
## ASSERTS ##
No. of searchspace entries where fractions do not sum to 100.0: 0
No. of searchspace entries that have duplicate solvent labels: 0
No. of searchspace entries with permutation-invariant combinations: 0
No. of unique 1-solvent entries (exp. 4) 4
No. of unique 2-solvent entries (exp. 18) 18
No. of unique 3-solvent entries (exp. 12) 12