Merge pull request #244 from alan-turing-institute/simplify

Simplify

autoemulate/compare.py

@@ -79,7 +79,7 @@ def setup(
         param_search : bool
             Whether to perform hyperparameter search over predifined parameter grids.
         param_search_type : str
-            Type of hyperparameter search to perform. Can be "grid", "random", or "bayes".
+            Type of hyperparameter search to perform. Currently only "random".
         param_search_iters : int
             Number of parameter settings that are sampled. Only used if
             param_search=True and param_search_type="random".

autoemulate/emulators/__init__.py

@@ -1,11 +1,10 @@
 from .conditional_neural_process import ConditionalNeuralProcess
-from .gaussian_process import GaussianProcess
 from .gaussian_process_mogp import GaussianProcessMOGP
+from .gaussian_process_sklearn import GaussianProcessSklearn
 from .gaussian_process_torch import GaussianProcessTorch
 from .gradient_boosting import GradientBoosting
 from .light_gbm import LightGBM
 from .neural_net_sk import NeuralNetSk
-from .neural_net_torch import NeuralNetTorch
 from .polynomials import SecondOrderPolynomial
 from .radial_basis_functions import RadialBasisFunctions
 from .random_forest import RandomForest
@@ -16,16 +15,12 @@
     RadialBasisFunctions().model_name: RadialBasisFunctions(),
     RandomForest().model_name: RandomForest(),
     GradientBoosting().model_name: GradientBoosting(),
-    GaussianProcess().model_name: GaussianProcess(),
-    SupportVectorMachines().model_name: SupportVectorMachines(),
     LightGBM().model_name: LightGBM(),
-    NeuralNetTorch(module="MultiLayerPerceptron").model_name: NeuralNetTorch(
-        module="MultiLayerPerceptron"
-    ),
-    NeuralNetTorch(module="RadialBasisFunctionsNetwork").model_name: NeuralNetTorch(
-        module="RadialBasisFunctionsNetwork"
-    ),
-    NeuralNetSk().model_name: NeuralNetSk(),
-    ConditionalNeuralProcess().model_name: ConditionalNeuralProcess(),
+    SupportVectorMachines().model_name: SupportVectorMachines(),
     GaussianProcessTorch().model_name: GaussianProcessTorch(),
+    ConditionalNeuralProcess().model_name: ConditionalNeuralProcess(),
+    # currently not in use
+    # NeuralNetSk().model_name: NeuralNetSk(),
+    # GaussianProcess().model_name: GaussianProcess(),
+    # GaussianProcessMOGP().model_name: GaussianProcessMOGP(),
 }

autoemulate/emulators/conditional_neural_process.py

@@ -232,6 +232,7 @@ def predict(self, X, return_std=False):
         # needs to be float64 to pass estimator tests
         # squeeze out batch dimension again so that score() etc. runs
         mean, logvar = predictions
+
         mean = mean[-X.shape[0] :].numpy().astype(np.float64).squeeze()
         logvar = logvar[-X.shape[0] :].numpy().astype(np.float64).squeeze()
 
@@ -241,10 +242,13 @@ def predict(self, X, return_std=False):
             var = np.exp(logvar) * (self._y_train_std**2)
             logvar = np.log(var)
 
-        # if y is 1d, make predictions same shape
+        # make sure predictions have the same shape as y
         if self.y_dim_ == 1:
             mean = mean.ravel()
             logvar = logvar.ravel()
+        else:
+            mean = mean.reshape(-1, self.output_dim_)
+            logvar = logvar.reshape(-1, self.output_dim_)
 
         if return_std:
             std = np.exp(0.5 * logvar)

autoemulate/emulators/gaussian_process.py → autoemulate/emulators/gaussian_process_sklearn.py

@@ -16,7 +16,7 @@
 from autoemulate.utils import _suppress_convergence_warnings
 
 
-class GaussianProcess(BaseEstimator, RegressorMixin):
+class GaussianProcessSklearn(BaseEstimator, RegressorMixin):
     """Gaussian Process Emulator.
 
     Wraps GaussianProcessRegressor from scikit-learn.

autoemulate/emulators/neural_networks/__init__.py

@@ -1,2 +0,0 @@
-from .base import TorchModule
-from .get_module import get_module