From f8531d7ee43ca74e9cc8690f276874bc66027a66 Mon Sep 17 00:00:00 2001
From: Maria Fernanda Morales <65073126+mfmo45@users.noreply.github.com>
Date: Thu, 8 Aug 2024 16:45:59 +0200
Subject: [PATCH] [PCA] Fixed PCA bootstrap loops
---
src/bayesvalidrox/surrogate_models/engine.py | 4 +++-
.../surrogate_models/polynomial_chaos.py | 18 ++++++++--------
.../surrogate_models/surrogate_models.py | 21 +++++++++----------
3 files changed, 22 insertions(+), 21 deletions(-)
diff --git a/src/bayesvalidrox/surrogate_models/engine.py b/src/bayesvalidrox/surrogate_models/engine.py
index d5110f94e..a77dff09b 100644
--- a/src/bayesvalidrox/surrogate_models/engine.py
+++ b/src/bayesvalidrox/surrogate_models/engine.py
@@ -127,7 +127,9 @@ class Engine:
# Prepare X samples
# For training the surrogate use ExpDesign.X_tr, ExpDesign.X is for the model to run on
- # TODO: set to some more general value, chech how much this is needed!
+ # TODO: set to some more general value, check how much this is needed!
+ # TODO: for GPs, the maxdeg should be set to 1, because self.emulator will also be True, or give the
+ # Parent class a _pce_deg=1 attribute automatically
if self.emulator:
maxdeg = np.max(MetaModel._pce_deg)
else:
diff --git a/src/bayesvalidrox/surrogate_models/polynomial_chaos.py b/src/bayesvalidrox/surrogate_models/polynomial_chaos.py
index 7f8c708ec..e83eb9ff1 100644
--- a/src/bayesvalidrox/surrogate_models/polynomial_chaos.py
+++ b/src/bayesvalidrox/surrogate_models/polynomial_chaos.py
@@ -295,6 +295,15 @@ class PCE(MetaModel):
range(output.shape[1]))
out = list(results)
+ # Store the first out dictionary
+ if self.fast_bootstrap and b_i == 0:
+ self.first_out[key] = copy.deepcopy(out)
+
+ # Update the coefficients with OLS during bootstrap-iters
+ if b_i > 0 and self.fast_bootstrap:
+ out = self.update_pce_coeffs(
+ X, output, self.first_out[key])
+
# Create a dict to pass the variables
for i in range(output.shape[1]):
self._deg_dict[f'b_{b_i + 1}'][key][f"y_{i + 1}"] = out[i]['degree']
@@ -306,15 +315,6 @@ class PCE(MetaModel):
# TODO: this is commented out here, but should be used in the SeqDesign??
# self._LCerror[f'b_{b_i+1}'][key][f"y_{i+1}"] = out[i]['_LCerror']
- # Store the first out dictionary
- if self.fast_bootstrap and b_i == 0:
- self.first_out[key] = copy.deepcopy(out)
-
- # Update the coefficients with OLS during bootstrap-iters
- if b_i > 0 and self.fast_bootstrap:
- out = self.update_pce_coeffs(
- X, output, self.first_out[key])
-
# -------------------------------------------------------------------------
def update_pce_coeffs(self, X, y, out_dict=None):
diff --git a/src/bayesvalidrox/surrogate_models/surrogate_models.py b/src/bayesvalidrox/surrogate_models/surrogate_models.py
index 94b3dd50f..d6c968944 100644
--- a/src/bayesvalidrox/surrogate_models/surrogate_models.py
+++ b/src/bayesvalidrox/surrogate_models/surrogate_models.py
@@ -231,7 +231,7 @@ def transform_y(self, y, b_i=0, trafo_type=''):
# Start transformation
pca, y_transform[key], n_comp = self.pca_transformation(
- y, self.n_pca_components)
+ y[key], self.n_pca_components)
self.pca[f'b_{b_i + 1}'][key] = pca
# Store the number of components for fast bootstrapping
@@ -303,16 +303,12 @@ def _bootstrap_eval(eval_function):
kwargs['b_i'] = b_i
mean_pred, std_pred = eval_function(self, *args, **kwargs)
- # Apply inverse transformations
- for i in range(mean_pred[list(mean_pred.keys())[0]].shape[1]):
- # Save predictions for each output
- if self.dim_red_method.lower() == 'pca':
- pca = self.pca[f'b_{b_i + 1}'][f"y_{i + 1}"]
- mean_pred[f"y_{i + 1}"] = pca.inverse_transform(mean_pred)
- std_pred[f"y_{i + 1}"] = np.zeros(mean_pred.shape)
- else:
- mean_pred[f"y_{i + 1}"] = mean_pred
- std_pred[f"y_{i + 1}"] = std_pred
+ # Appy inverse transformation
+ if self.dim_red_method.lower() == 'pca':
+ for output, values in mean_pred.items():
+ pca = self.pca[f'b_{b_i + 1}'][output]
+ mean_pred[output] = pca.inverse_transform(values)
+ std_pred[output] = np.zeros(values.shape)
# Save predictions for each bootstrap iteration
mean_pred_b[b_i] = mean_pred
@@ -487,6 +483,9 @@ class MetaModel:
The number of features is set by `self.n_pca_components`.
If this is not given, `self.var_pca_threshold` is used as a threshold.
+ ToDo: Check the inputs needed for this class, there is an error when PCA is used.
+ ToDo: From the y_transformation() function, a dictionary is being sent instead of an array for target.
+
Parameters
----------
target : array of shape (n_samples,)
--
GitLab