Fbopt matchduva

domainlab/algos/builder_matchdg.py

@@ -62,6 +62,8 @@ def init_business(self, exp):
             i_c=task.isize.i_c,
             i_h=task.isize.i_h,
             i_w=task.isize.i_w)
+        # different than model, ctr_model has no classification so it has
+        # different wrapper
         ctr_model = ModelWrapMatchDGNet(ctr_net, list_str_y=task.list_str_y)
         ctr_model = ctr_model.to(device)
 

domainlab/models/a_model_classif.py

@@ -60,6 +60,13 @@ def evaluate(self, loader_te, device):
         logger = Logger.get_logger()
         logger.info(f"before training, model accuracy: {acc}")
 
+    def extract_semantic_feat(self, tensor_x):
+        """
+        by default, use the logit as extracted feature if the current method
+        is not being overriden by child class
+        """
+        return self.cal_logit_y(tensor_x)
+
     @abc.abstractmethod
     def cal_logit_y(self, tensor_x):
         """
@@ -199,4 +206,4 @@ def cal_reg_loss(self, tensor_x, tensor_y, tensor_d, others=None):
         """
         device = tensor_x.device
         bsize = tensor_x.shape[0]
-        return [torch.zeros(bsize, 1).to(device)], [0.0]
+        return [torch.zeros(bsize, 1).to(device)], [0.0]

test_match_duva.sh

@@ -0,0 +1,6 @@
+python main_out.py --te_d 0 1 2 --tr_d 3 7 --task=mnistcolor10 --debug --bs=2 --aname=matchhduva \
+  --epochs_ctr=3 --epos=6 --nname=conv_bn_pool_2 --gamma_y=7e5 \
+        --nname_topic_distrib_img2topic=conv_bn_pool_2 \
+        --nname_encoder_sandwich_layer_img2h4zd=conv_bn_pool_2
+
+

test_match_duva_vlcs.sh

@@ -0,0 +1,4 @@
+python main_out.py --te_d=caltech --task=mini_vlcs --debug --bs=2 --aname=matchhduva \
+  --epochs_ctr=3 --epos=6 --npath=examples/nets/resnet.py --gamma_y=7e5 \
+        --npath_topic_distrib_img2topic=examples/nets/resnet.py \
+        --npath_encoder_sandwich_layer_img2h4zd=examples/nets/resnet.py