Possible to test another validation set that have labels diffierent from train set and test set?

[andyhx]

hi,I want to test another dataset(assuming A) when training the network, I want to callucate the loss and accurate rate of the model,this dateset is different labels from trainset and testset ,it is not serapated from the dateset divided into two trainset and testset
so is it right to just add this dataset A into the trainval.prototxt by adding the test stages in the phase: TEST, or i have to modify the caffe testnet code to test A set and rebuild the caffe to do this ,so is there a more simple way or an example to follow ,thank you in avdance!

[andyhx]

anyone knows , I have been google for a long time ,but I cant find any cues

[andyhx]

any guide?

[zhaifly]

did you use the new test dataset(A) to training the network?
or you want to use a trained network to test your new dataset?
what the difference between the two lables?

[andyhx]

hi,@zhaifly ,thanks for ur reply,
1.I dont uset the new test dataset(A) to train the network
2.Yes I want to test the A dataset's FRR and ERR result , when training not after the train finished ;
3.the A lables is index from 0-100,train and test is 0-2000,their index is referece to anthoer persons ,It is a new dataset ,not same as train and test dataset!

[zhaifly]

@andyhx is there has any relation between labels(0100) and labels(02000), if has, i think you can map the labels(0100) to labels(02000) manually. if not, i think you need to modify your network and use your new data to train it and then test it.

i am new to caffe and DL, may it help you!

[andyhx]

@zhaifly thank u,there is no connections between them,I think training a new data is not a good idea,So
anyone has come across the same problem,help me ,thank u!