diff --git a/datasets/eval_query.txt b/datasets/eval_query.txt
index cb6a97d..fa57bcc 100644
--- a/datasets/eval_query.txt
+++ b/datasets/eval_query.txt
@@ -16,4 +16,4 @@
 10000129_1_CTce_ThAb.mat
 10000136_1_CTce_ThAb.mat
 10000110_1_CTce_ThAb.mat
-10000130_1_CTce_ThAb.mat
+10000105_1_CTce_ThAb.mat
diff --git a/datasets/eval_support.txt b/datasets/eval_support.txt
index 6de01b3..ea11171 100644
--- a/datasets/eval_support.txt
+++ b/datasets/eval_support.txt
@@ -1 +1 @@
-10000105_1_CTce_ThAb.mat
\ No newline at end of file
+10000130_1_CTce_ThAb.mat
diff --git a/run_oneshot.py b/run_oneshot.py
index e91a993..64b38c3 100644
--- a/run_oneshot.py
+++ b/run_oneshot.py
@@ -47,7 +47,7 @@ def train(train_params, common_params, data_params, net_params):
     train_data, test_data = load_data(data_params)
 
     folds = ['fold1']
-    model_prefix = 'model6_Focal_loss_noClsLastDec_'
+    model_prefix = 'model6_Focal_gamma2_after3epoch_'
     for fold in folds:
         final_model_path = os.path.join(common_params['save_model_dir'], model_prefix + fold + '.pth.tar')
 
@@ -121,7 +121,7 @@ def evaluate(eval_params, net_params, data_params, common_params, train_params):
     model_name = 'model6_Dice_L2_loss_target_fold1.pth.tar'
     folds = ['fold1']
 
-    eval_model_path1 = "saved_models/model6_Dice_L2_loss_target_fold1.pth.tar"
+    eval_model_path1 = "saved_models/model6_Focal_gamma2_after3epoch_fold1.pth.tar"
     eval_model_path2 = "saved_models/model6_coronal_fold1.pth.tar"
     eval_model_path3 = "saved_models/model6_sagittal_fold1.pth.tar"
 
diff --git a/settings.ini b/settings.ini
index 472ae27..f41d3af 100644
--- a/settings.ini
+++ b/settings.ini
@@ -37,7 +37,7 @@ exp_name = "model6_fold1"
 final_model_file = "model6_fold1_DiceL2.pth.tar"
 learning_rate = 1e-1
 momentum = 0.95
-train_batch_size = 6
+train_batch_size = 8
 val_batch_size = 5
 log_nth = 10
 num_epochs = 20
@@ -64,4 +64,4 @@ support_txt_file = "datasets/eval_support.txt"
 remap_config = "WholeBody"
 #Valid options : COR, AXI, SAG
 orientation = "AXI"
-save_predictions_dir = "predictions_1view_DiceL2_target"
\ No newline at end of file
+save_predictions_dir = "predictions_Focal_gamma2_after3epoch"
\ No newline at end of file
diff --git a/solver_oneshot_multiOpti_auto.py b/solver_oneshot_multiOpti_auto.py
index 3b731c8..3db1ff6 100644
--- a/solver_oneshot_multiOpti_auto.py
+++ b/solver_oneshot_multiOpti_auto.py
@@ -50,11 +50,11 @@ def __init__(self,
         # self.optim = optim(model.parameters(), **optim_args)
 
         self.optim_c = optim(
-            [{'params': model.conditioner.parameters(), 'lr': 1e-15, 'momentum': 0.95, 'weight_decay': 0.001}
+            [{'params': model.conditioner.parameters(), 'lr': 1e-2, 'momentum': 0.95, 'weight_decay': 0.0001}
              ], **optim_args)
 
         self.optim_s = optim(
-            [{'params': model.segmentor.parameters(), 'lr': 1e-15, 'momentum': 0.95, 'weight_decay': 0.001}
+            [{'params': model.segmentor.parameters(), 'lr': 1e-2, 'momentum': 0.95, 'weight_decay': 0.0001}
              ], **optim_args)
 
         # self.scheduler = lr_scheduler.StepLR(self.optim, step_size=5,
@@ -153,7 +153,7 @@ def train(self, train_loader, test_loader):
 
                     if model.is_cuda:
                         condition_input, query_input, y2, y1 = condition_input.cuda(self.device,
-                                                                                non_blocking=True), query_input.cuda(
+                                                                                    non_blocking=True), query_input.cuda(
                             self.device,
                             non_blocking=True), y2.cuda(
                             self.device, non_blocking=True), y1.cuda(
@@ -165,13 +165,14 @@ def train(self, train_loader, test_loader):
                     # space_w, channel_w = weights
                     # e_w1, e_w2, e_w3, bn_w, d_w3, d_w2, d_w1, cls_w = space_w
                     # e_c1, e_c2, e_c3, bn_c, d_c3, d_c2, d_c1, cls_c = channel_w
-                    e_w1, e_w2, e_w3, bn_w, d_w3, d_w2, d_w1, cls_w = weights
-                    weights = [e_w1, e_w2, e_w3, bn_w, d_w3, d_w2, None, None]
+                    # e_w1, e_w2, e_w3, bn_w, d_w3, d_w2, d_w1, cls_w = weights
+                    # weights = [e_w1, e_w2, e_w3, bn_w, d_w3, d_w2, d_w1, cls_w]
                     # channel_w = [e_c1, e_c2, e_c3, bn_c, d_c3, d_c2, d_c1, cls_c]
                     # weights = (space_w, channel_w)
                     output = model.segmentor(query_input, weights)
                     # TODO: add weights
-                    loss = self.loss_func(F.softmax(output, dim=1), y2, y1)
+                    cost_weight = (1, 0) if epoch < 3 else (0, 1)
+                    loss = self.loss_func(F.softmax(output, dim=1), y2, weight=cost_weight)
                     optim_s.zero_grad()
                     optim_c.zero_grad()
                     loss.backward()