Mse

metrics/MSE.py

@@ -0,0 +1,11 @@
+import numpy as np
+import random
+import logging
+
+def mse(X, obs):
+    '''
+    Retourne la MSE
+    '''
+    random_idx = random.sample(range(X.shape[0]), 1)[0]   
+    _mse = np.nanmean(((X[random_idx] - obs.squeeze())**2),axis=(-2,-1))
+    return _mse

metrics/__init__.py

@@ -19,6 +19,7 @@
 from metrics import CRPS_calc
 from metrics import area_proportion as ap
 from metrics import object_detection as obj
+from metrics import MSE
 
 from metrics.metrics import Metric, PreprocessCondObs, PreprocessDist, PreprocessStandalone
 
@@ -474,6 +475,25 @@ def _calculateCore(self, processed_data):
 
         return GM.relative_std_diff(real_data,fake_data)
 
+
+#####################################################################
+############################ Determinstic metrics ###################
+#####################################################################
+
+class mse(PreprocessCondObs):
+    def __init__(self, *args, **kwargs):
+        super().__init__(isBatched=True)
+
+    def _calculateCore(self, processed_data):
+        if not self.isOnReal:
+            exp_data = processed_data['fake_data']
+        else:
+            exp_data = processed_data['real_data']
+        obs_data = processed_data['obs_data']
+        return MSE.mse(exp_data, obs_data)
+
+
+
 #####################################################################
         ######################################################
 #####################################################################

metrics/wind_comp.py

@@ -113,7 +113,8 @@ def computeWindDir(U, V, xRef=None, yRef=None, proj=None):
     @rtype: tuple
     @return: vitesse (m/s) et direction du vent exprimée en degrés météo (0 = vent du nord).
     """
-    ff = np.sqrt(U * U + V * V)
+
+    ff = np.sqrt(np.maximum(U * U + V * V, np.zeros_like(U)))
 
     dd3 = (180 + 180 / np.pi * np.arctan2(U, V)) % 360
     # logging.debug(dd3)

preprocess/rrPreprocessor.py

@@ -38,6 +38,7 @@ def init_normalization(self):
         normalization_type = self.normalization["type"]
         if normalization_type == "mean":
             means, stds = self.load_stat_files(normalization_type, "mean", "std")
+            logging.debug(f"stat constants {means, stds}")
             return None, None, means, stds
         elif normalization_type == "minmax":
             maxs, mins = self.load_stat_files(normalization_type, "max", "min")
@@ -64,16 +65,16 @@ def load_stat_files(self, normalization_type, str1, str2):
             std_or_min_filename += "_ppx"
         mean_or_max_filename += ".npy"
         std_or_min_filename += ".npy"
-        logging.debug(f"{mean_or_max_filename}", f"{std_or_min_filename}")
-        logging.debug(f"Normalization set to {normalization_type}")
+        # logging.debug(f"{mean_or_max_filename}", f"{std_or_min_filename}")
+        # logging.debug(f"Normalization set to {normalization_type}")
         stat_folder = self.config_data["stat_folder"]
         file_path = os.path.join(self.config_data["real_data_dir"], stat_folder, mean_or_max_filename)
         means_or_maxs = np.load(file_path).astype('float32')
-        logging.debug(f"{str1} file found, {means_or_maxs.shape}")
+        # logging.debug(f"{str1} file found, {means_or_maxs.shape}")
 
         file_path = os.path.join(self.config_data["real_data_dir"], stat_folder, std_or_min_filename)
         stds_or_mins = np.load(file_path).astype('float32')
-        logging.debug(f"{str2} file found, {stds_or_mins.shape}")
+        # logging.debug(f"{str2} file found, {stds_or_mins.shape}")
         return means_or_maxs, stds_or_mins
 
     def detransform(self, data):
@@ -178,4 +179,5 @@ def __init__(self, config_data, sizeH, sizeW, variables, **kwargs):
         super().__init__(config_data, sizeH, sizeW, variables, **kwargs)
 
     def process_batch(self, batch):
-        return self.detransform(batch)
+        res = self.detransform(batch)
+        return res

requirements.txt

@@ -1,5 +1,5 @@
 PyYaml==6.0.1
-torch==2.2.0+cu121
+torch==2.2.0
 numpy==1.26.4
 tqdm==4.66.1
 pandas==2.2.0
@@ -9,4 +9,5 @@ pyproj==3.6.1
 properscoring==0.1
 geopy==2.4.1
 astropy==6.0.0
+scikit-image
 CRPS