Bayesian Exploration with NaNs¶

As violations of constraints can lead to invalid values of the objective, the evaluate function may simply return NaNs. We demonstrate below how we can still perform Bayesian Exploration in that case.

In [1]:

# set values if testing
import os

import pandas as pd
import torch
from copy import deepcopy
from xopt import Xopt, Evaluator
from xopt.generators.bayesian import BayesianExplorationGenerator
from xopt.resources.test_functions.tnk import evaluate_TNK, tnk_vocs

from gest_api.vocs import GreaterThanConstraint, LessThanConstraint

# Ignore all warnings
import warnings

warnings.filterwarnings("ignore")

SMOKE_TEST = os.environ.get("SMOKE_TEST")
NUM_MC_SAMPLES = 1 if SMOKE_TEST else 128
NUM_RESTARTS = 1 if SMOKE_TEST else 20

vocs = deepcopy(tnk_vocs)
vocs.objectives = {"y1": "EXPLORE"}

# set values if testing import os import pandas as pd import torch from copy import deepcopy from xopt import Xopt, Evaluator from xopt.generators.bayesian import BayesianExplorationGenerator from xopt.resources.test_functions.tnk import evaluate_TNK, tnk_vocs from gest_api.vocs import GreaterThanConstraint, LessThanConstraint # Ignore all warnings import warnings warnings.filterwarnings("ignore") SMOKE_TEST = os.environ.get("SMOKE_TEST") NUM_MC_SAMPLES = 1 if SMOKE_TEST else 128 NUM_RESTARTS = 1 if SMOKE_TEST else 20 vocs = deepcopy(tnk_vocs) vocs.objectives = {"y1": "EXPLORE"}

/home/runner/work/Xopt/Xopt/.venv/lib/python3.12/site-packages/pyro/ops/stats.py:527: SyntaxWarning: invalid escape sequence '\g'
  we have :math:`ES^{*}(P,Q) \ge ES^{*}(Q,Q)` with equality holding if and only if :math:`P=Q`, i.e.

In [2]:

# modify the evaluate function to return NaNs if constraints are violated
def evaluate(input_dict):
    output_dict = evaluate_TNK(input_dict)
    del output_dict["y2"]
    for c in vocs.constraints.keys():
        if (
            isinstance(vocs.constraints[c], GreaterThanConstraint)
            and output_dict[c] <= vocs.constraints[c].value
        ):
            output_dict["y1"] = torch.nan
        elif (
            isinstance(vocs.constraints[c], LessThanConstraint)
            and output_dict[c] >= vocs.constraints[c].value
        ):
            output_dict["y1"] = torch.nan
    return output_dict

# modify the evaluate function to return NaNs if constraints are violated def evaluate(input_dict): output_dict = evaluate_TNK(input_dict) del output_dict["y2"] for c in vocs.constraints.keys(): if ( isinstance(vocs.constraints[c], GreaterThanConstraint) and output_dict[c] <= vocs.constraints[c].value ): output_dict["y1"] = torch.nan elif ( isinstance(vocs.constraints[c], LessThanConstraint) and output_dict[c] >= vocs.constraints[c].value ): output_dict["y1"] = torch.nan return output_dict

In [3]:

generator = BayesianExplorationGenerator(vocs=vocs)
generator.max_travel_distances = [0.25, 0.25]
generator.n_monte_carlo_samples = NUM_MC_SAMPLES
generator.numerical_optimizer.n_restarts = NUM_RESTARTS

evaluator = Evaluator(function=evaluate)
X = Xopt(generator=generator, evaluator=evaluator)
X

generator = BayesianExplorationGenerator(vocs=vocs) generator.max_travel_distances = [0.25, 0.25] generator.n_monte_carlo_samples = NUM_MC_SAMPLES generator.numerical_optimizer.n_restarts = NUM_RESTARTS evaluator = Evaluator(function=evaluate) X = Xopt(generator=generator, evaluator=evaluator) X

Out[3]:

            Xopt
________________________________
Version: 0.1.dev1+gca942dd06
Data size: 0
Config as YAML:
dump_file: null
evaluator:
  function: __main__.evaluate
  function_kwargs: {}
  max_workers: 1
  vectorized: false
generator:
  computation_time: null
  custom_objective: null
  fixed_features: null
  gp_constructor:
    covar_modules: {}
    custom_noise_prior: null
    mean_modules: {}
    name: standard
    train_config: null
    train_kwargs: null
    train_method: lbfgs
    train_model: true
    trainable_mean_keys: []
    transform_inputs: true
    use_cached_hyperparameters: false
    use_low_noise_prior: false
  max_travel_distances:
  - 0.25
  - 0.25
  model: null
  n_candidates: 1
  n_interpolate_points: null
  n_monte_carlo_samples: 128
  name: bayesian_exploration
  numerical_optimizer:
    max_iter: 2000
    max_time: 5.0
    n_restarts: 20
    name: LBFGS
  returns_id: false
  supports_batch_generation: true
  supports_constraints: true
  supports_multi_objective: true
  supports_single_objective: true
  turbo_controller: null
  use_cuda: false
  vocs:
    constants:
      a:
        dtype: null
        type: Constant
        value: dummy_constant
    constraints:
      c1:
        dtype: null
        type: GreaterThanConstraint
        value: 0.0
      c2:
        dtype: null
        type: LessThanConstraint
        value: 0.5
    objectives:
      y1:
        dtype: null
        type: ExploreObjective
    observables: {}
    variables:
      x1:
        default_value: null
        domain:
        - 0.0
        - 3.14159
        dtype: null
        type: ContinuousVariable
      x2:
        default_value: null
        domain:
        - 0.0
        - 3.14159
        dtype: null
        type: ContinuousVariable
serialize_inline: false
serialize_torch: false
stopping_condition: null
strict: true

Run exploration¶

We start with evaluating 2 points that we know satisfy the constraints. We then run 30 exploration steps.

In [4]:

X.evaluate_data(pd.DataFrame({"x1": [1.0, 0.75], "x2": [0.7, 0.95]}))

X.evaluate_data(pd.DataFrame({"x1": [1.0, 0.75], "x2": [0.7, 0.95]}))

Out[4]:

	x1	x2	a	y1	c1	c2	xopt_runtime	xopt_error
0	1.00	0.70	dummy_constant	1.00	0.584045	0.290	0.002593	False
1	0.75	0.95	dummy_constant	0.75	0.494833	0.265	0.000144	False

In [5]:

N_STEPS = 1 if SMOKE_TEST else 30
for i in range(N_STEPS):
    print(f"step {i}")
    X.step()

N_STEPS = 1 if SMOKE_TEST else 30 for i in range(N_STEPS): print(f"step {i}") X.step()

step 0

step 1

step 2

step 3

step 4

step 5

step 6

step 7

step 8

step 9

step 10

step 11

step 12

step 13

step 14

step 15

step 16

step 17

step 18

step 19

step 20

step 21

step 22

step 23

step 24

step 25

step 26

step 27

step 28

step 29

In [6]:

# view the data
X.data

# view the data X.data

Out[6]:

	x1	x2	a	y1	c1	c2	xopt_runtime	xopt_error
0	1.000000	0.700000	dummy_constant	1.000000	0.584045	0.290000	0.002593	False
1	0.750000	0.950000	dummy_constant	0.750000	0.494833	0.265000	0.000144	False
2	0.000000	1.735397	dummy_constant	NaN	1.911604	1.776207	0.004596	False
3	0.663353	1.058575	dummy_constant	0.663353	0.649854	0.338690	0.004417	False
4	1.006497	1.535983	dummy_constant	NaN	2.471253	1.329800	0.008819	False
5	1.612694	0.984256	dummy_constant	NaN	2.648709	1.472591	0.002675	False
6	0.827296	0.385012	dummy_constant	NaN	-0.244725	0.120345	0.000170	False
7	0.358040	0.984898	dummy_constant	0.358040	0.022013	0.255279	0.000169	False
8	1.013053	0.200759	dummy_constant	1.013053	0.166575	0.352768	0.000166	False
9	1.176503	0.476586	dummy_constant	1.176503	0.512073	0.458204	0.000167	False
10	0.840300	0.000000	dummy_constant	NaN	-0.393895	0.365804	0.000165	False
11	1.188028	0.514192	dummy_constant	1.188028	0.578971	0.473584	0.000167	False
12	0.504238	0.969011	dummy_constant	0.504238	0.175621	0.219990	0.000170	False
13	0.095290	1.030575	dummy_constant	0.095290	0.061621	0.445301	0.000167	False
14	0.284031	1.129254	dummy_constant	0.284031	0.425489	0.442603	0.000173	False
15	1.000088	0.934860	dummy_constant	1.000088	0.788326	0.439192	0.000164	False
16	1.095606	0.165936	dummy_constant	1.095606	0.301965	0.466346	0.000165	False
17	1.881004	0.000000	dummy_constant	NaN	2.438175	2.157172	0.000164	False
18	1.196125	0.537824	dummy_constant	1.196125	0.631162	0.486020	0.000164	False
19	0.410727	0.982412	dummy_constant	0.410727	0.033970	0.240691	0.000184	False
20	0.045319	1.028828	dummy_constant	NaN	-0.015662	0.486394	0.000163	False
21	0.054663	1.040623	dummy_constant	0.054663	0.019115	0.490598	0.000166	False
22	0.672815	1.162888	dummy_constant	0.672815	0.856254	0.469286	0.000165	False
23	1.016347	0.960334	dummy_constant	1.016347	0.865300	0.478522	0.000165	False
24	0.234387	1.139353	dummy_constant	0.234387	0.452515	0.479323	0.000171	False
25	1.015534	0.968339	dummy_constant	1.015534	0.876144	0.485117	0.000167	False
26	0.230137	0.182941	dummy_constant	NaN	-0.888890	0.173353	0.000169	False
27	0.000000	0.000000	dummy_constant	NaN	-1.100000	0.500000	0.000166	False
28	0.746910	0.785397	dummy_constant	0.746910	0.082687	0.142416	0.000169	False
29	1.034880	0.061852	dummy_constant	1.034880	0.017053	0.478070	0.000166	False
30	1.196963	0.445347	dummy_constant	1.196963	0.547632	0.488744	0.000167	False
31	0.585893	1.190184	dummy_constant	0.585893	0.708837	0.483732	0.000165	False

In [7]:

# plot results
ax = X.data.plot("x1", "x2")
ax.set_aspect("equal")

# plot results ax = X.data.plot("x1", "x2") ax.set_aspect("equal")