def transIndex(S,SlicedData,modelPath,rootdir,outdir,SliceSize=64,SlidingWindow=64,frameNo=10,depthDirFromRoot=0): [model, _, _] = LOADModel(modelPath) mergedirs=[dirs for dirs,_,_ in os.walk(rootdir) if len(dirs.split("\\"))==(len(rootdir.split("\\"))+depthDirFromRoot)] for dir in mergedirs: print(f"\n--------------Inside {dir}-----------------\n") savedir=outdir+'\\'+os.path.basename(Path(dir)) os.makedirs(savedir,exist_ok=True) temp=np.array(list(map(lambda x: np.array(x[1:],dtype="int") if dir in x[0].decode()[4:] else None,SlicedData.F))) temp=np.array(list(filter(lambda x: x is not None, temp))) Bothslices=np.array([],dtype="int64") Startslices=np.array([],dtype="int64") Endslices=np.array([],dtype="int64") Allslices=np.array([],dtype="int64") i=1 for file in temp: start=int((file[0]-SliceSize)/SlidingWindow)+1 end=int((file[1]-SliceSize)/SlidingWindow)+1 Y=np.array(SlicedData.Y[file[0]:file[1]]).reshape(-1) n=int((Y.shape[0]-SliceSize)/SlidingWindow)+1 Y=np.reshape(Y, (n, SliceSize)) Y = np.array([[np.zeros((1),dtype="int16"), np.where(np.diff(i) != 0)[0] + 1,start+ind] if len(np.unique(i, return_counts=True)[1]) == 1 else [np.ones((1),dtype="int16"), np.where(np.diff(i) != 0)[0] + 1,start+ind] for ind,i in enumerate(Y)]) bothslices = np.array([ i[2] for i in Y if (i[1] == frameNo).any() and (SliceSize - i[1] == frameNo).any()],dtype="int64") endslices= np.array([ i[2] for i in Y if (SliceSize - i[1] == frameNo).any()],dtype="int64") startslices= np.array([ i[2] for i in Y if (i[1] == frameNo).any()],dtype="int64") Bothslices=np.append(Bothslices,bothslices) Startslices=np.append(Startslices,startslices) Endslices=np.append(Endslices,endslices) Allslices=np.append(Allslices,Y[:,2].astype("int64")) if i%150==0: print(f"----------------{i} of {temp.shape[0]} files processed-------------") i+=1 print(f"\n------------Started Prediction for all transition slices {len(Allslices)}-------------\n") X=SlicedData.Xs[Allslices] Y_pred=model.predict(X[:,:,1:],batch_size=32,verbose=0) Y_pred=np.argmax(Y_pred,axis=1) Y=np.argmax(SlicedData.Ys[Allslices],axis=1) confusionAll=metrics.confusion_matrix(Y,Y_pred,labels=[0,1]) _,recallAll,_,_=metrics.precision_recall_fscore_support(Y,Y_pred,labels=[0,1]) if Startslices.size!=0: print(f"\n------------Started Prediction for beginning transition slices {len(Startslices)}-------------\n") X=SlicedData.Xs[Startslices] Y_pred=model.predict(X[:,:,1:],batch_size=32,verbose=0) Y_pred=np.argmax(Y_pred,axis=1) Y=np.argmax(SlicedData.Ys[Startslices],axis=1) confusionStart=metrics.confusion_matrix(Y,Y_pred,labels=[0,1]) _,recallStart,_,_=metrics.precision_recall_fscore_support(Y,Y_pred,labels=[0,1]) errorS10=(confusionStart[1,0]/confusionAll[1,0])*100 errorS01=(confusionStart[0,1]/confusionAll[0,1])*100 if Endslices.size!=0: print(f"\n------------Started Prediction for ending transition slices {len(Endslices)}-------------\n") X=SlicedData.Xs[Endslices] Y_pred=model.predict(X[:,:,1:],batch_size=32,verbose=0) Y_pred=np.argmax(Y_pred,axis=1) Y=np.argmax(SlicedData.Ys[Endslices],axis=1) confusionEnd=metrics.confusion_matrix(Y,Y_pred,labels=[0,1]) _,recallEnd,_,_=metrics.precision_recall_fscore_support(Y,Y_pred,labels=[0,1]) errorE10=(confusionEnd[1,0]/confusionAll[1,0] )*100 errorE01=(confusionEnd[0,1]/confusionAll[0,1] )*100 if Bothslices.size!=0: print(f"\n------------Started Prediction for transition slices at both start & end {len(Bothslices)}-------------\n") X=SlicedData.Xs[Bothslices] Y_pred=model.predict(X[:,:,1:],batch_size=32,verbose=0) Y_pred=np.argmax(Y_pred,axis=1) Y=np.argmax(SlicedData.Ys[Bothslices],axis=1) confusionBoth=metrics.confusion_matrix(Y,Y_pred,labels=[0,1]) _,recallBoth,_,_=metrics.precision_recall_fscore_support(Y,Y_pred,labels=[0,1]) errorB10=(confusionBoth[1,0]/confusionAll[1,0] )*100 errorB01=(confusionBoth[0,1]/confusionAll[0,1] )*100 with open(f"{savedir}\\Confusion_Matrix_transitionAt_{frameNo}.csv", 'w') as f: np.savetxt(f, [f"{dir}"], fmt="%s", delimiter=",") if Startslices.size!=0: np.savetxt(f, [f"Error transit/no-transit pos={frameNo}th frame "], fmt="%s", delimiter=",") np.savetxt(f, [f"{confusionStart[1,0]}/{confusionAll[1,0]}"], fmt="%s", delimiter=",") np.savetxt(f, [errorS10], fmt="%s", delimiter=",") np.savetxt(f, [f"Error no-transit/transit pos={frameNo}th frame "], fmt="%s", delimiter=",") np.savetxt(f, [f"{confusionStart[0,1]}/{confusionAll[0,1]}"], fmt="%s", delimiter=",") np.savetxt(f, [errorS01], fmt="%s", delimiter=",") np.savetxt(f, [f"Total slices having transition at pos={frameNo}th frame"], fmt="%s", delimiter=",") np.savetxt(f, [f"{Startslices.size}"], fmt="%s", delimiter=",") if Endslices.size!=0: np.savetxt(f, [f"Error transit/no-transit pos={SliceSize-frameNo}th frame "], fmt="%s", delimiter=",") np.savetxt(f, [f"{confusionEnd[1,0]}/{confusionAll[1,0]}"], fmt="%s", delimiter=",") np.savetxt(f, [errorE10], fmt="%s", delimiter=",") np.savetxt(f, [f"Error no-transit/transit pos={SliceSize-frameNo}th frame "], fmt="%s", delimiter=",") np.savetxt(f, [f"{confusionEnd[0,1]}/{confusionAll[0,1]}"], fmt="%s", delimiter=",") np.savetxt(f, [errorE01], fmt="%s", delimiter=",") np.savetxt(f, [f"Total slices having transition at pos={SliceSize-frameNo}th frame"], fmt="%s", delimiter=",") np.savetxt(f, [f"{Endslices.size}"], fmt="%s", delimiter=",") if Bothslices.size!=0: np.savetxt(f, ["Error transit/no-transit pos=both "], fmt="%s", delimiter=",") np.savetxt(f, [f"{confusionBoth[1,0]}/{confusionAll[1,0]}"], fmt="%s", delimiter=",") np.savetxt(f, [errorB10], fmt="%s", delimiter=",") np.savetxt(f, ["Error no-transit/transit pos=both "], fmt="%s", delimiter=",") np.savetxt(f, [f"{confusionBoth[0,1]}/{confusionAll[0,1]}"], fmt="%s", delimiter=",") np.savetxt(f, [errorB01], fmt="%s", delimiter=",") np.savetxt(f, [f"Total slices having transition at both start/end positions "], fmt="%s", delimiter=",") np.savetxt(f, [f"{Bothslices.size}"], fmt="%s", delimiter=",") print("\n------------Done--------------\n") return