I want to know that during the process of comparing one deep neural network model for two different data sets , is it valid to perform fine tuning of a model. Example:
For first data sets, the embedding size = 128, Convolutional layer filter size = 2 and batch size =32 but For second data set after tuning the parameters, embedding size = 32, Convolutional layer filter size = 4 and batch size = 64.
means can i use different values of parameter of the same model for different data sets.Is it logical comparison of the same model for the different data sets?
Dear Anam Habib ,
This is a normal behavior. Since different datasets have different data distributions the hyperparameters found (filter size, embedding size) can vary. Therefore, your comparison is not wrong. However, I think you need to define better your problem/goal. For example, you can have two main goals:
Hello Anam
This depends on the definition and aim of your problem.
If you want to evaluate a single architecture in terms of accuracy, time, etc., the parameters need to be constant and identical.
but, If you want to achieve best result and accuracy for each dataset, you can evaluate different architectures with different parameters. In this case, the obtained architecture may not be optimal in terms of other features, such as computational time, etc.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs