go down to working levels

Blunde1 · May 15, 2024 · a31ae59 · a31ae59
1 parent 281c0cf
commit a31ae59
Showing 1 changed file with 10 additions and 9 deletions.
diff --git a/src/ert/analysis/_es_update.py b/src/ert/analysis/_es_update.py
@@ -34,6 +34,7 @@
 from iterative_ensemble_smoother.experimental import (
     AdaptiveESMDA,
 )
+from sklearn.preprocessing import StandardScaler
 from typing_extensions import Self
 
 from ..config.analysis_module import ESSettings, IESSettings
@@ -667,8 +668,8 @@ def adaptive_localization_progress_callback(
                 source, iens_active_index, param_group
             )
             X_local = temp_storage[param_group]
-            # X_local_scaler = StandardScaler()
-            # X_scaled = X_local_scaler.fit_transform(X_local.T)
+            X_local_scaler = StandardScaler()
+            X_scaled = X_local_scaler.fit_transform(X_local.T)
             # scaler_cache[param_group] = X_local_scaler
 
             graph_u_sub = config_node.load_parameter_graph(
@@ -683,13 +684,13 @@ def adaptive_localization_progress_callback(
             Z = sp.sparse.csc_matrix(Z)
 
             Prec_u_sub = gspme.prec_sparse(
-                X_local.T,
+                X_scaled,
                 Z,
                 markov_order=1,
                 cov_shrinkage=True,
                 symmetrization=False,
                 shrinkage_target=2,
-                inflation_factor=15.0,
+                inflation_factor=10.0,
             )
 
             # # A very simple hash key for graph
@@ -770,13 +771,13 @@ def adaptive_localization_progress_callback(
         ) as file:
             pickle.dump(observation_errors, file)
 
-        # X_full_scaler = StandardScaler()
-        # X_full_scaled = X_full_scaler.fit_transform(X_full.T)
-        # print(f"Scaled X_full has shape: {X_full_scaled.shape}")
+        X_full_scaler = StandardScaler()
+        X_full_scaled = X_full_scaler.fit_transform(X_full.T)
+        print(f"Scaled X_full has shape: {X_full_scaled.shape}")
 
         # Call fit: Learn sparse linear map only
         H = linear_boost_ic_regression(
-            U=X_full.T,
+            U=X_full_scaled,
             Y=S.T,
             learning_rate=0.95,
             effective_dimension=0,
@@ -803,7 +804,7 @@ def adaptive_localization_progress_callback(
         # Call transport? might have to do some coding here
         # Perhaps use an iterative solver instead of direct spsolve or similar
         X_full = gtmap.transport(
-            X_full.T,
+            X_full_scaled,
             S.T,
             observation_values,
             update_indices=update_indices,