Consolidate ert parameters precision to 32bits

 - Give a precision interval in test
 - update snapshots

Author: xjules

src/ert/analysis/_es_update.py

@@ -299,8 +299,8 @@ def adaptive_localization_progress_callback(
         np.fill_diagonal(T, T.diagonal() + 1)
 
     def correlation_callback(
-        cross_correlations_of_batch: npt.NDArray[np.float64],
-        cross_correlations_accumulator: list[npt.NDArray[np.float64]],
+        cross_correlations_of_batch: npt.NDArray[np.float32],
+        cross_correlations_accumulator: list[npt.NDArray[np.float32]],
     ) -> None:
         cross_correlations_accumulator.append(cross_correlations_of_batch)
 
@@ -350,7 +350,7 @@ def correlation_callback(
             progress_callback(AnalysisStatusEvent(msg=log_msg))
 
             start = time.time()
-            cross_correlations: list[npt.NDArray[np.float64]] = []
+            cross_correlations: list[npt.NDArray[np.float32]] = []
             for param_batch_idx in batches:
                 update_idx = param_batch_idx[non_zero_variance_mask[param_batch_idx]]
                 X_local = param_ensemble_array[update_idx, :]

src/ert/analysis/_update_commons.py

@@ -96,7 +96,7 @@ def _auto_scale_observations(
     obs_mask: npt.NDArray[np.bool_],
     active_realizations: list[str],
     progress_callback: Callable[[AnalysisEvent], None],
-) -> tuple[npt.NDArray[np.float64], pl.DataFrame] | tuple[None, None]:
+) -> tuple[npt.NDArray[np.float32], pl.DataFrame] | tuple[None, None]:
     """
     Performs 'Auto Scaling' to mitigate issues with correlated observations,
     and saves computed scaling factors across input groups to ERT storage.
@@ -328,7 +328,7 @@ class _OutlierColumns(StrEnum):
 def _all_parameters(
     ensemble: Ensemble,
     iens_active_index: npt.NDArray[np.int_],
-) -> npt.NDArray[np.float64]:
+) -> npt.NDArray[np.float32]:
     """Return all parameters in assimilation problem"""
 
     param_groups = list(ensemble.experiment.parameter_configuration.keys())

src/ert/analysis/misfit_preprocessor.py

@@ -29,7 +29,7 @@ def get_scaling_factor(nr_observations: int, nr_components: int) -> float:
 
 
 def get_nr_primary_components(
-    responses: npt.NDArray[np.float64], threshold: float
+    responses: npt.NDArray[np.float32], threshold: float
 ) -> int:
     """
     Calculate the number of principal components required
@@ -61,7 +61,7 @@ def get_nr_primary_components(
 
 
 def cluster_responses(
-    responses: npt.NDArray[np.float64],
+    responses: npt.NDArray[np.float32],
     nr_clusters: int,
 ) -> npt.NDArray[np.int_]:
     """
@@ -70,7 +70,7 @@ def cluster_responses(
     be clustered together.
     """
     correlation = spearmanr(responses).statistic
-    if isinstance(correlation, np.float64):
+    if isinstance(correlation, np.floating):
         correlation = np.array([[1, correlation], [correlation, 1]])
     # Take absolute value to cluster based on correlation strength rather
     # than direction.
@@ -84,9 +84,9 @@ def cluster_responses(
 
 
 def main(
-    responses: npt.NDArray[np.float64],
-    obs_errors: npt.NDArray[np.float64],
-) -> tuple[npt.NDArray[np.float64], npt.NDArray[np.int_], npt.NDArray[np.int_]]:
+    responses: npt.NDArray[np.floating],
+    obs_errors: npt.NDArray[np.floating],
+) -> tuple[npt.NDArray[np.floating], npt.NDArray[np.int_], npt.NDArray[np.int_]]:
     """
     Perform 'Auto Scaling' to mitigate issues with correlated observations in ensemble
     smoothers.

src/ert/config/design_matrix.py

@@ -391,14 +391,14 @@ def convert_numeric_string_columns(df: pl.DataFrame) -> pl.DataFrame:
     for col, dtype in zip(df.columns, df.dtypes, strict=False):
         if dtype == pl.String:
             try:
-                df = df.with_columns(pl.col(col).cast(pl.Int64, strict=True).alias(col))
+                df = df.with_columns(pl.col(col).cast(pl.Int32, strict=True).alias(col))
                 continue
             except InvalidOperationError:
                 pass
 
             try:  # noqa: SIM105
                 df = df.with_columns(
-                    pl.col(col).cast(pl.Float64, strict=True).alias(col)
+                    pl.col(col).cast(pl.Float32, strict=True).alias(col)
                 )
             except InvalidOperationError:
                 pass

src/ert/config/ext_param_config.py

@@ -78,7 +78,7 @@ def write_to_runpath(
 
     def create_storage_datasets(
         self,
-        from_data: npt.NDArray[np.float64],
+        from_data: npt.NDArray[np.float32],
         iens_active_index: npt.NDArray[np.int_],
     ) -> Iterator[tuple[int, xr.Dataset]]:
         for i, realization in enumerate(iens_active_index):
@@ -97,7 +97,7 @@ def create_storage_datasets(
 
     def load_parameters(
         self, ensemble: Ensemble, realizations: npt.NDArray[np.int_]
-    ) -> npt.NDArray[np.float64]:
+    ) -> npt.NDArray[np.float32]:
         raise NotImplementedError
 
     def load_parameter_graph(self) -> nx.Graph[int]:

src/ert/config/field.py

@@ -319,7 +319,7 @@ def write_to_runpath(
 
     def create_storage_datasets(
         self,
-        from_data: npt.NDArray[np.float64],
+        from_data: npt.NDArray[np.float32],
         iens_active_index: npt.NDArray[np.int_],
     ) -> Iterator[tuple[int, xr.Dataset]]:
         for i, realization in enumerate(iens_active_index):
@@ -336,7 +336,7 @@ def create_storage_datasets(
 
     def load_parameters(
         self, ensemble: Ensemble, realizations: npt.NDArray[np.int_]
-    ) -> npt.NDArray[np.float64]:
+    ) -> npt.NDArray[np.float32]:
         ds = ensemble.load_parameters(self.name, realizations)
         assert isinstance(ds, xr.Dataset)
         ensemble_size = len(ds.realizations)

src/ert/config/gen_kw_config.py

@@ -232,7 +232,7 @@ def write_to_runpath(
 
     def load_parameters(
         self, ensemble: Ensemble, realizations: npt.NDArray[np.int_]
-    ) -> npt.NDArray[np.float64]:
+    ) -> npt.NDArray[np.float32]:
         return (
             ensemble.load_parameters(self.name, realizations)
             .drop("realization")
@@ -242,7 +242,7 @@ def load_parameters(
 
     def create_storage_datasets(
         self,
-        from_data: npt.NDArray[np.float64],
+        from_data: npt.NDArray[np.float32],
         iens_active_index: npt.NDArray[np.int_],
     ) -> Iterator[tuple[int | None, pl.DataFrame]]:
         yield (

src/ert/config/parameter_config.py

@@ -100,7 +100,7 @@ def write_to_runpath(
     @abstractmethod
     def create_storage_datasets(
         self,
-        from_data: npt.NDArray[np.float64],
+        from_data: npt.NDArray[np.float32],
         iens_active_index: npt.NDArray[np.int_],
     ) -> Iterator[tuple[int | None, pl.DataFrame | xr.Dataset]]:
         """
@@ -128,7 +128,7 @@ def copy_parameters(
     @abstractmethod
     def load_parameters(
         self, ensemble: Ensemble, realizations: npt.NDArray[np.int_]
-    ) -> npt.NDArray[np.float64]:
+    ) -> npt.NDArray[np.float32]:
         """
         Load the parameter from internal storage for the given ensemble.
         Must return array of shape (number of parameters, number of realizations).

src/ert/config/surface_config.py

@@ -199,7 +199,7 @@ def write_to_runpath(
 
     def create_storage_datasets(
         self,
-        from_data: npt.NDArray[np.float64],
+        from_data: npt.NDArray[np.float32],
         iens_active_index: npt.NDArray[np.int_],
     ) -> Iterator[tuple[int, xr.Dataset]]:
         for i, realization in enumerate(iens_active_index):
@@ -229,7 +229,7 @@ def create_storage_datasets(
 
     def load_parameters(
         self, ensemble: Ensemble, realizations: npt.NDArray[np.int_]
-    ) -> npt.NDArray[np.float64]:
+    ) -> npt.NDArray[np.float32]:
         ds = ensemble.load_parameters(self.name, realizations)
         assert isinstance(ds, xr.Dataset)
         ensemble_size = len(ds.realizations)

src/ert/dark_storage/compute/misfits.py

@@ -6,9 +6,9 @@
 
 
 def _calculate_misfit(
-    obs_value: npt.NDArray[np.float64],
-    response_value: npt.NDArray[np.float64],
-    obs_std: npt.NDArray[np.float64],
+    obs_value: npt.NDArray[np.float32],
+    response_value: npt.NDArray[np.float32],
+    obs_std: npt.NDArray[np.float32],
 ) -> list[float]:
     difference = response_value - obs_value
     misfit = (difference / obs_std) ** 2