when to use ML and when to DL ?
When is it recommended to use deep learning for supervised algorithms (such classification) instead of using ordinary machine learning (such as xgboost) ?
Saying that deep learning is used for complex problematics is subjective. what are the signs of complexity ? number of rows ?.... from which we should start considered deep learning models
Saying that deep learning is used for complex problematics is subjective. what are the signs of complexity ? number of rows ?.... from which we should start considered deep learning models
1 answers ( 0 marked as helpful)
Hi,
As you already outlined the general advice that is given is to use deep learning for more complex problems, but that is very subjective.
Unfortunately, there is no one aspect that I can tell you to look for when deciding which method to use. Most of it will come with more experience, as you see different problems and how the differing methods perform on them.
What I will say is that in general, you want to use the simplest tool available that performs adequetly. For example, in certain situations a single decision tree may have very close performance to a random forest (collection of decision trees). In that case, you should use the decision tree as it provides great interpretability.
So, when presented with a problem, you can begin an exploratory analysis by starting with the simplest model (that makes sense with the given dataset/goal). E.g. that could be a linear regression. If this simple model does not give satisfactory results, you can switch out for a more complex model. Repeat that until you are satisfied with the initial outcome. Then, you can start optimizing that model more thoroughly.
If you do this in several different situations, you will begin to have intuition about the different ML approaches.
Hope this helps!
Best,
Nikola, 365 Team
As you already outlined the general advice that is given is to use deep learning for more complex problems, but that is very subjective.
Unfortunately, there is no one aspect that I can tell you to look for when deciding which method to use. Most of it will come with more experience, as you see different problems and how the differing methods perform on them.
What I will say is that in general, you want to use the simplest tool available that performs adequetly. For example, in certain situations a single decision tree may have very close performance to a random forest (collection of decision trees). In that case, you should use the decision tree as it provides great interpretability.
So, when presented with a problem, you can begin an exploratory analysis by starting with the simplest model (that makes sense with the given dataset/goal). E.g. that could be a linear regression. If this simple model does not give satisfactory results, you can switch out for a more complex model. Repeat that until you are satisfied with the initial outcome. Then, you can start optimizing that model more thoroughly.
If you do this in several different situations, you will begin to have intuition about the different ML approaches.
Hope this helps!
Best,
Nikola, 365 Team