diff --git a/climax/main/config.py b/climax/main/config.py
index 2ab3734969a1e0df3f0990bce2dd882aa279e224..a86c8d9ad6e80ba2d0b2ab6400ac18f99502a704 100644
--- a/climax/main/config.py
+++ b/climax/main/config.py
@@ -50,7 +50,7 @@ CHUNKS = {'time': 365}
 # -----------------------------------------------------------------------------
 
 # include day of year as predictor
-DOY = False
+DOY = True
 
 # use digital elevation model instead of model orography
 DEM = True
@@ -85,7 +85,7 @@ STRATIFY = False
 # size of the validation set w.r.t. the training set
 # e.g., VALID_SIZE = 0.1 means: 90% of CALIB_PERIOD for training
 #                               10% of CALIB_PERIOD for validation
-VALID_SIZE = 0.1
+VALID_SIZE = 0.2
 
 # number of folds for training with KFold cross-validation
 CV = 5
@@ -117,18 +117,18 @@ FILTERS = [32, 64, 128, 256]
 #     BernoulliGammaLoss (NLL of Bernoulli-Gamma distribution)
 #     BernoulliWeibullLoss (NLL of Bernoulli-Weibull distribution)
 # LOSS = L1Loss()
-LOSS = MSELoss()
-# LOSS = BernoulliGammaLoss(min_amount=1)
+# LOSS = MSELoss()
+LOSS = BernoulliGammaLoss(min_amount=1)
 # LOSS = BernoulliWeibullLoss(min_amount=1)
 
 # stochastic optimization algorithm
 OPTIM = torch.optim.SGD
 # OPTIM = torch.optim.Adam
-OPTIM_PARAMS = {'lr': 1e-1, # learning rate
-                'weight_decay': 1e-6  # regularization rate
+OPTIM_PARAMS = {'lr': 1e-3, # learning rate
+                'weight_decay': 0  # regularization rate
                 }
 if OPTIM == torch.optim.SGD:
-    OPTIM_PARAMS['momentum'] = 0.9
+    OPTIM_PARAMS['momentum'] = 0.99
 
 # learning rate scheduler
 # LR_SCHEDULER = torch.optim.lr_scheduler.MultiStepLR
@@ -148,14 +148,14 @@ BATCH_SIZE = 16
 # network training configuration
 TRAIN_CONFIG = {
     'checkpoint_state': {},
-    'epochs': 250,
+    'epochs': 50,
     'save': True,
     'save_loaders': False,
     'early_stop': True,
-    'patience': 25,
+    'patience': 10,
     'multi_gpu': True,
     'classification': False,
-    'clip_gradients': True
+    'clip_gradients': False
     }
 
 # whether to overwrite existing models