Hello, during this week I learnt about training set, cv set and test set.
If you have a neural model with a lot of units and a model with minimum units, as the image indicates. What does it mean? The sentence is clear, but if you have a cv set with more error than training that means you need you modify your model?
What is an error value that you can say: this is Ok?
Thanks.
Gus
Hi @gmazzaglia this overall means that your model is not able to generalize to unseen data, there are several approach you could take, that can go from changing your model to optimize your model and add things to your data. Complex models tends to overfit more than simple models, my usual approach is to go with simple model and try to identify areas to improve, for instance, if adding a layer makes it a better model on validation data, I might try with more layers and see how that goes but you should always watch out with overfit.
Let me know if this helps
This could mean your model is too complex, and has overfit the training set.
You’re looking for a balanced relationship between the two metrics. Not necessarily identical, just “close enough”.
There are no numerical thresholds on this, it’s a judgement call based on your experience and skill.
Thanks, @pastorsoto and @TMosh.
Regards.
Gus
