The subject as above, I see a lot of people this question, but there is no solution, how to solve?

CodePudding user response:

Because you did not give a concrete model description, I can only give you some general advice
This case suggests that your model without training, not in school, the parameter is not updated, common reasons:
1. The unreasonable loss function definition
2. Decreasing process optimization algorithm is not suitable, stuck in the local optimum or saddle surface
Vector (3) set up not appropriate, too big or too small may be, the possibility of too small a few bigger
4. Training data quantity is too little, or the training sample convergence
5. Batch size is not appropriate

CodePudding user response:

reference 1st floor sigtem response:

because you did not give a specific description model, I can only give you some general advice
This case suggests that your model without training, not in school, the parameter is not updated, common reasons:
1. The unreasonable loss function definition
2. Decreasing process optimization algorithm is not suitable, stuck in the local optimum or saddle surface
Vector (3) set up not appropriate, too big or too small may be, the possibility of too small a few bigger
4. Training data quantity is too little, or the training sample convergence
5. Batch size inappropriate

You get me wrong, I mean the training and testing the custom on the sensitivity and specificity of two values is always the same, but each round will be changed,