Fix filter must sum=0

.gitignore

@@ -7,4 +7,4 @@ Dataset/
 .vscode
 /*/__pycache__
 .mypy_cache
-/FilterEvaluator/image*
+/*/image*

FilterEvaluator/Evaluator.py

@@ -71,26 +71,27 @@ traindata, testdata = Loader.Cifar10Mono(batchsize, num_workers=0, shuffle=False
 
 
 
-for batch_idx, (data, target) in enumerate(traindata):
-    utils.NumpyToImage(data.cpu().detach().numpy(), CurrentPath+"image", title="TrainData")
-    break
+# for batch_idx, (data, target) in enumerate(traindata):
+#     utils.NumpyToImage(data.cpu().detach().numpy(), CurrentPath+"image", title="TrainData")
+#     break
 
-weight,active = EvaluatorUnsuper.UnsuperLearnSearchWeight(model, layer, traindata, NumSearch=1,SaveChannel=8,SearchChannelRatio=1, Interation=128)
-utils.NumpyToImage(weight, CurrentPath+"image",title="SearchWeight")
-
-b =0
-
-
-
-
-
-
-
-
-# weight,active = EvaluatorUnsuper.UnsuperLearnSearchWeight(model, layer, traindata, NumSearch=100000, SearchChannelRatio=32, Interation=10)
-# np.save("WeightSearch.npy", weight)
-# weight = np.load(CurrentPath+"WeightSearch.npy")
+# weight, active = EvaluatorUnsuper.UnsuperLearnSearchWeight(
+#     model, layer, traindata, NumSearch=10, SaveChannel=4000, SearchChannelRatio=32, Interation=10)
 # utils.NumpyToImage(weight, CurrentPath+"image",title="SearchWeight")
+
+# b =0
+
+
+
+
+
+
+
+
+weight,active = EvaluatorUnsuper.UnsuperLearnSearchWeight(model, layer, traindata, NumSearch=100000, SearchChannelRatio=32, Interation=10)
+np.save("WeightSearch.npy", weight)
+weight = np.load(CurrentPath+"WeightSearch.npy")
+utils.NumpyToImage(weight, CurrentPath+"image",title="SearchWeight")
 # weight = np.load(CurrentPath+"WeightSearch.npy")
 # bestweight,index = EvaluatorUnsuper.UnsuperLearnFindBestWeight(model,layer,weight,traindata,128,100000)
 # np.save(CurrentPath+"bestweightSearch.npy", bestweight)

FilterEvaluator/EvaluatorUnsuper.py

@@ -53,6 +53,9 @@ def UnsuperLearnSearchWeight(model, layer, dataloader, NumSearch=10000, SaveChan
     minactive = np.empty((0))
     minweight = np.empty([0,newweightshape[-3],newweightshape[-2],newweightshape[-1]])
     newweight = np.random.uniform(-1.0,1.0,newweightshape).astype("float32")
+    newweight = newweight.reshape((-1,newweightshape[-1]*newweightshape[-2]))
+    newweight = np.swapaxes(newweight,0,1)-np.mean(newweight,-1)
+    newweight = np.swapaxes(newweight,0,1).reshape(newweightshape)
 
     dataset = []
     for batch_idx, (data, target) in enumerate(dataloader):