Hi, Andrew said in course 3 that we have to train the model keeping in mind our target and target is training+evaluation. What if the model is already trained and deployed on the remote. do we need to change the model every time and train the model for different evaluation if comes in future or it can be done dynamically? As far as i can think is we need to train the model again and that can take time so if there is an API that gives evaluation parameters in request the response time will be much higher.
Welcome to the community!
Sorry for the late response. Well, it depends on the model you are trying to train. If the model has the same usage and properties, it can be used to train other models keeping in mind how you are going to train them. Transfer learning is just one phenomenon.
For instance, if you have a model trained for bird detection in the sky, you can certainly train it on other objects for the same purpose and goal like chasing firefly etc. The test images then would be accountable here.
Note: training a model carries high computational costs and its time consuming too.