I mean what is determine my model size, connection amount between layers and neurons, or size of my dataset?


Dataset and model refer to different things. Dataset means part of data available for training (training dataset) or validation (validation dataset). Model is the learning process goal, the state of the computer "brain" after it has been fully educated (or made its learning). Model size refers to size of the container which contains the model. In deep learning it can be measured by width and depth of the network used, I also found a site comparing different models by npy file size, that physically contains the generated model as computer code. In that case model contained a more complex structure which was documented and size in bytes was for comparison purposes.

So in short, it is roughly speaking the size of layers and neurons, if I have to take one of your options. Dataset is a different thing.

More precise explanation about what is model and what is dataset:


  • $\begingroup$ Thank you for your answer I understand that size of dataset doesn't effect the size of model. For example I have keras sequential model with 3 layers of Dense() with 64 units each. This amount of layers and units detmermine the size of model and this size is not depend on what size of dataset I use (1 GB or 50GB) right? $\endgroup$
    – T K
    Aug 22 '19 at 6:58
  • $\begingroup$ That is my understanding. The model is more accurate the more you train it, but that does not add the size, at least on Deep Learning where I am familiar with. $\endgroup$
    – mico
    Aug 22 '19 at 8:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.