Skip to content

Time Dependent Model Constructor

Bases: StandardModelConstructor

Time-dependent model constructor for Bayesian optimization.

Attributes:

Name Type Description
name str

The name of the model constructor.
use_spectral_mixture_kernel bool

Whether to use the Spectral Mixture Kernel for the time axis.
initialize_spectral_kernel_from_data bool

Whether to initialize the Spectral Mixture Kernel from data.

Methods:

Name Description
build_model

Build the model.
build_model_from_vocs

Build the model from VOCS.
Source code in xopt/generators/bayesian/models/time_dependent.py 
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
class TimeDependentModelConstructor(StandardModelConstructor):
    """
    Time-dependent model constructor for Bayesian optimization.

    Attributes
    ----------
    name : str
        The name of the model constructor.
    use_spectral_mixture_kernel : bool
        Whether to use the Spectral Mixture Kernel for the time axis.
    initialize_spectral_kernel_from_data : bool
        Whether to initialize the Spectral Mixture Kernel from data.

    Methods
    -------
    build_model(self, input_names: List[str], outcome_names: List[str], data: pd.DataFrame, input_bounds: Dict[str, List] = None, dtype: torch.dtype = torch.double, device: Union[torch.device, str] = "cpu") -> ModelListGP
        Build the model.
    build_model_from_vocs(self, vocs: VOCS, data: pd.DataFrame, dtype: torch.dtype = torch.double, device: Union[torch.device, str] = "cpu") -> ModelListGP
        Build the model from VOCS.
    """

    name: str = Field("time_dependent", frozen=True)
    use_spectral_mixture_kernel: bool = True
    initialize_spectral_kernel_from_data: bool = False

    def build_model(
        self,
        input_names: List[str],
        outcome_names: List[str],
        data: pd.DataFrame,
        input_bounds: Dict[str, List] = None,
        dtype: torch.dtype = torch.double,
        device: Union[torch.device, str] = "cpu",
    ) -> ModelListGP:
        """
        Build the model.

        Parameters
        ----------
        input_names : List[str]
            The names of the input variables.
        outcome_names : List[str]
            The names of the outcome variables.
        data : pd.DataFrame
            The data to use for building the model.
        input_bounds : Dict[str, List], optional
            The bounds for the input variables. Defaults to None.
        dtype : torch.dtype, optional
            The data type to use. Defaults to torch.double.
        device : Union[torch.device, str], optional
            The device to use. Defaults to "cpu".

        Returns
        -------
        ModelListGP
            The built model.
        """
        # get model input names
        new_input_names = deepcopy(input_names)
        new_input_names += ["time"]

        min_t = data["time"].min()
        max_t = data["time"].max() + 15.0
        new_input_bounds = deepcopy(input_bounds)
        new_input_bounds["time"] = [min_t, max_t]

        # set covar modules if not specified -- use SpectralMixtureKernel for time axis
        # see Kuklev, N., et al. "Online accelerator tuning with adaptive
        # bayesian optimization." Proc. NAPAC 22 (2022): 842.
        if self.use_spectral_mixture_kernel:
            covar_modules = {}
            for name in outcome_names:
                if len(input_names) == 1:
                    matern_dims = [0]
                else:
                    matern_dims = tuple(range(len(input_names)))
                time_dim = [len(input_names)]

                matern_kernel = MaternKernel(
                    nu=2.5,
                    active_dims=matern_dims,
                    lengthscale_prior=GammaPrior(3.0, 6.0),
                )
                spectral_kernel = SpectralMixtureKernel(
                    num_mixtures=3, active_dims=time_dim
                )

                if self.initialize_spectral_kernel_from_data:
                    train_X, train_Y, train_Yvar = get_training_data(
                        new_input_names, name, data
                    )

                    # can only initialize spectral kernel from data if there are
                    # more than one training data point
                    if len(train_X) > 1:
                        spectral_kernel.initialize_from_data(train_X, train_Y)
                    else:
                        raise RuntimeWarning(
                            "cannot initialize spectral kernel from a "
                            "single data sample, may negatively impact"
                            " performance"
                        )

                covar_modules[name] = ProductKernel(spectral_kernel, matern_kernel)

            self.covar_modules = covar_modules

        return super().build_model(
            new_input_names, outcome_names, data, new_input_bounds, dtype, device
        )

    def build_model_from_vocs(
        self,
        vocs: VOCS,
        data: pd.DataFrame,
        dtype: torch.dtype = torch.double,
        device: Union[torch.device, str] = "cpu",
    ) -> ModelListGP:
        """
        Build the model from VOCS.

        Parameters
        ----------
        vocs : VOCS
            The VOCS object containing the variables, objectives, and constraints.
        data : pd.DataFrame
            The data to use for building the model.
        dtype : torch.dtype, optional
            The data type to use. Defaults to torch.double.
        device : Union[torch.device, str], optional
            The device to use. Defaults to "cpu".

        Returns
        -------
        ModelListGP
            The built model.
        """
        return self.build_model(
            vocs.variable_names + ["time"],
            vocs.output_names,
            data,
            {
                n: v.domain if isinstance(v, ContinuousVariable) else v
                for n, v in vocs.variables.items()
            },
            dtype,
            device,
        )

build_heteroskedastic_gp(X, Y, Yvar, train=True, **kwargs) staticmethod

Utility method for creating and training heteroskedastic SingleTaskGP models.

Parameters:

Name Type Description Default
X Tensor

Training data for input variables.

 required
Y Tensor

Training data for outcome variables.

 required
Yvar Tensor

Training data for outcome variable variances.

 required
train (bool, True)

Flag to specify if hyperparameter training should take place

 True
**kwargs

Additional keyword arguments for model configuration.

 {}

Returns:

Type Description
Model

The trained heteroskedastic SingleTaskGP model.
Notes

Heteroskedastic modeling incurs a number of warnings from botorch, which are suppressed within this method.

Source code in xopt/generators/bayesian/base_model.py 
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
@staticmethod
def build_heteroskedastic_gp(
    X: Tensor, Y: Tensor, Yvar: Tensor, train: bool = True, **kwargs
) -> Model:
    """
    Utility method for creating and training heteroskedastic SingleTaskGP models.

    Parameters
    ----------
    X : Tensor
        Training data for input variables.
    Y : Tensor
        Training data for outcome variables.
    Yvar : Tensor
        Training data for outcome variable variances.
    train : bool, True
        Flag to specify if hyperparameter training should take place
    **kwargs
        Additional keyword arguments for model configuration.

    Returns
    -------
    Model
        The trained heteroskedastic SingleTaskGP model.

    Notes
    -----
    Heteroskedastic modeling incurs a number of warnings from botorch, which are
    suppressed within this method.

    """
    warnings.warn(
        "Heteroskedastic modeling has been removed from botorch due "
        "to numerical stability issues. A copy of the implementation "
        "is included in Xopt, however it may be unstable / buggy. "
        "Your results may vary and keep an eye on warnings."
    )

    if X.shape[0] == 0 or Y.shape[0] == 0 or Yvar.shape[0] == 0:
        raise ValueError("no data found to train model!")

    with warnings.catch_warnings():
        warnings.filterwarnings("ignore")
        model = XoptHeteroskedasticSingleTaskGP(X, Y, Yvar, **kwargs)

    if train:
        try:
            with warnings.catch_warnings():
                warnings.filterwarnings("ignore")
                mll = ExactMarginalLogLikelihood(model.likelihood, model)
                fit_gpytorch_mll(mll)
        except ModelFittingError:
            warnings.warn(
                "Model fitting failed for heteroskedastic GP. Returning untrained model."
            )
    return model

build_mean_module(name, mean_modules, input_transform, outcome_transform)

Build the mean module for the output specified by name.

Parameters:

Name Type Description Default
name str

The name of the output.

 required
mean_modules

The dictionary of mean modules.

 required
input_transform InputTransform

Transform for input variables.

 required
outcome_transform OutcomeTransform

Transform for outcome variables.

 required

Returns:

Type Description
Optional[CustomMean]

The mean module for the output, or None if not specified.
Source code in xopt/generators/bayesian/models/standard.py 
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
def build_mean_module(
    self, name, mean_modules, input_transform, outcome_transform
) -> Optional[CustomMean]:
    """
    Build the mean module for the output specified by name.

    Parameters
    ----------
    name : str
        The name of the output.
    mean_modules: dict
        The dictionary of mean modules.
    input_transform : InputTransform
        Transform for input variables.
    outcome_transform : OutcomeTransform
        Transform for outcome variables.

    Returns
    -------
    Optional[CustomMean]
        The mean module for the output, or None if not specified.

    """
    mean_module = self._get_module(mean_modules, name)
    if mean_module is not None:
        fixed_model = False if name in self.trainable_mean_keys else True
        mean_module = CustomMean(
            mean_module, input_transform, outcome_transform, fixed_model=fixed_model
        )
    return mean_module

build_model(input_names, outcome_names, data, input_bounds=None, dtype=torch.double, device='cpu')

Build the model.

Parameters:

Name Type Description Default
input_names List[str]

The names of the input variables.

 required
outcome_names List[str]

The names of the outcome variables.

 required
data DataFrame

The data to use for building the model.

 required
input_bounds Dict[str, List]

The bounds for the input variables. Defaults to None.

 None
dtype dtype

The data type to use. Defaults to torch.double.

 double
device Union[device, str]

The device to use. Defaults to "cpu".

 'cpu'

Returns:

Type Description
ModelListGP

The built model.
Source code in xopt/generators/bayesian/models/time_dependent.py 
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
def build_model(
    self,
    input_names: List[str],
    outcome_names: List[str],
    data: pd.DataFrame,
    input_bounds: Dict[str, List] = None,
    dtype: torch.dtype = torch.double,
    device: Union[torch.device, str] = "cpu",
) -> ModelListGP:
    """
    Build the model.

    Parameters
    ----------
    input_names : List[str]
        The names of the input variables.
    outcome_names : List[str]
        The names of the outcome variables.
    data : pd.DataFrame
        The data to use for building the model.
    input_bounds : Dict[str, List], optional
        The bounds for the input variables. Defaults to None.
    dtype : torch.dtype, optional
        The data type to use. Defaults to torch.double.
    device : Union[torch.device, str], optional
        The device to use. Defaults to "cpu".

    Returns
    -------
    ModelListGP
        The built model.
    """
    # get model input names
    new_input_names = deepcopy(input_names)
    new_input_names += ["time"]

    min_t = data["time"].min()
    max_t = data["time"].max() + 15.0
    new_input_bounds = deepcopy(input_bounds)
    new_input_bounds["time"] = [min_t, max_t]

    # set covar modules if not specified -- use SpectralMixtureKernel for time axis
    # see Kuklev, N., et al. "Online accelerator tuning with adaptive
    # bayesian optimization." Proc. NAPAC 22 (2022): 842.
    if self.use_spectral_mixture_kernel:
        covar_modules = {}
        for name in outcome_names:
            if len(input_names) == 1:
                matern_dims = [0]
            else:
                matern_dims = tuple(range(len(input_names)))
            time_dim = [len(input_names)]

            matern_kernel = MaternKernel(
                nu=2.5,
                active_dims=matern_dims,
                lengthscale_prior=GammaPrior(3.0, 6.0),
            )
            spectral_kernel = SpectralMixtureKernel(
                num_mixtures=3, active_dims=time_dim
            )

            if self.initialize_spectral_kernel_from_data:
                train_X, train_Y, train_Yvar = get_training_data(
                    new_input_names, name, data
                )

                # can only initialize spectral kernel from data if there are
                # more than one training data point
                if len(train_X) > 1:
                    spectral_kernel.initialize_from_data(train_X, train_Y)
                else:
                    raise RuntimeWarning(
                        "cannot initialize spectral kernel from a "
                        "single data sample, may negatively impact"
                        " performance"
                    )

            covar_modules[name] = ProductKernel(spectral_kernel, matern_kernel)

        self.covar_modules = covar_modules

    return super().build_model(
        new_input_names, outcome_names, data, new_input_bounds, dtype, device
    )

build_model_from_vocs(vocs, data, dtype=torch.double, device='cpu')

Build the model from VOCS.

Parameters:

Name Type Description Default
vocs VOCS

The VOCS object containing the variables, objectives, and constraints.

 required
data DataFrame

The data to use for building the model.

 required
dtype dtype

The data type to use. Defaults to torch.double.

 double
device Union[device, str]

The device to use. Defaults to "cpu".

 'cpu'

Returns:

Type Description
ModelListGP

The built model.
Source code in xopt/generators/bayesian/models/time_dependent.py 
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
def build_model_from_vocs(
    self,
    vocs: VOCS,
    data: pd.DataFrame,
    dtype: torch.dtype = torch.double,
    device: Union[torch.device, str] = "cpu",
) -> ModelListGP:
    """
    Build the model from VOCS.

    Parameters
    ----------
    vocs : VOCS
        The VOCS object containing the variables, objectives, and constraints.
    data : pd.DataFrame
        The data to use for building the model.
    dtype : torch.dtype, optional
        The data type to use. Defaults to torch.double.
    device : Union[torch.device, str], optional
        The device to use. Defaults to "cpu".

    Returns
    -------
    ModelListGP
        The built model.
    """
    return self.build_model(
        vocs.variable_names + ["time"],
        vocs.output_names,
        data,
        {
            n: v.domain if isinstance(v, ContinuousVariable) else v
            for n, v in vocs.variables.items()
        },
        dtype,
        device,
    )

build_single_task_gp(X, Y, train=True, **kwargs) staticmethod

Utility method for creating and training simple SingleTaskGP models.

Parameters:

Name Type Description Default
X Tensor

Training data for input variables.

 required
Y Tensor

Training data for outcome variables.

 required
train (bool, True)

Flag to specify if hyperparameter training should take place

 True
**kwargs

Additional keyword arguments for model configuration.

 {}

Returns:

Type Description
Model

The trained SingleTaskGP model.
Source code in xopt/generators/bayesian/base_model.py 
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
@staticmethod
def build_single_task_gp(X: Tensor, Y: Tensor, train=True, **kwargs) -> Model:
    """
    Utility method for creating and training simple SingleTaskGP models.

    Parameters
    ----------
    X : Tensor
        Training data for input variables.
    Y : Tensor
        Training data for outcome variables.
    train : bool, True
        Flag to specify if hyperparameter training should take place
    **kwargs
        Additional keyword arguments for model configuration.

    Returns
    -------
    Model
        The trained SingleTaskGP model.

    """
    if X.shape[0] == 0 or Y.shape[0] == 0:
        raise ValueError("no data found to train model!")
    model = SingleTaskGP(X, Y, **kwargs)

    if train:
        mll = ExactMarginalLogLikelihood(model.likelihood, model)
        fit_gpytorch_mll(mll)
    return model

get_likelihood(batch_shape=torch.Size())

Get the likelihood for the model, considering the low noise prior and or a custom noise prior.

Returns:

Type Description
Likelihood

The likelihood for the model.
Source code in xopt/generators/bayesian/models/standard.py 
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
def get_likelihood(
    self,
    batch_shape: torch.Size = torch.Size(),
) -> Likelihood:
    """
    Get the likelihood for the model, considering the low noise prior and or a
    custom noise prior.

    Returns
    -------
    Likelihood
        The likelihood for the model.

    """
    if self.custom_noise_prior is not None:
        likelihood = GaussianLikelihood(
            noise_prior=self.custom_noise_prior, batch_shape=batch_shape
        )
    elif self.use_low_noise_prior:
        likelihood = GaussianLikelihood(
            noise_prior=GammaPrior(1.0, 100.0), batch_shape=batch_shape
        )
    else:
        noise_prior = GammaPrior(1.1, 0.05)
        noise_prior_mode = (noise_prior.concentration - 1) / noise_prior.rate
        likelihood = GaussianLikelihood(
            noise_prior=noise_prior,
            noise_constraint=GreaterThan(
                MIN_INFERRED_NOISE_LEVEL,
                transform=None,
                initial_value=noise_prior_mode,
            ),
            batch_shape=batch_shape,
        )
    return likelihood

yaml(**kwargs)

serialize first then dump to yaml string

Source code in xopt/pydantic.py 
231
232
233
234
235
236
237
238
def yaml(self, **kwargs):
    """serialize first then dump to yaml string"""
    output = json.loads(
        self.to_json(
            **kwargs,
        )
    )
    return yaml.dump(output)