Working on predictive model and getting the following metric results:
Cross-validation scores: [-1.10058189 -1.13730722 -1.1263573 -0.0854548 -0.9
9912831]
Average cross-validation score -0.8897659034581814
pipeline created
Train RMSE: 9.102893360089265
Test RMSE: 50123440.58756547
Train R^2: 0.9793272628026964
Test R^2: -416657966845.5751
The train r^2 score is very high, but the test data seems to be overfitting, what’s the best approach in resolving this?
Hi, Alf.
You have filed this under “General Discussion”, so it’s not possible (for me at least) to tell which course you are asking about. You’ll generally have better luck getting the attention of the people who know the topic area in question if you file your questions in the relevant category and subcategory. You can actually move the thread as described on this FAQ Topic.
Hi Paul,
This isn’t associated with any course unfortunately, just a side project i’m working on and need help with. What chat or discussion would best fit a question such as this?
If it’s not a course specific question, then I guess General Discussion is as good as any. But in that case, then you’ll have to give us a bit more context. I’ve only taken DLS and GANs and in those contexts, I’ve never heard the term R score. So perhaps all that means is that I’m not the right person to answer your question in the first place and we’ll need to wait until someone with the appropriate knowledge notices.
But just looking at this from a “pure math” perspective: whatever R is, how could R^2 be negative?
Well, unless you’re dealing with complex numbers here, but I’ve never heard of using anything other than \mathbb{R}^n in ML contexts …
But here again, this probably just shows that you need the attention of someone with the relevant knowledge 
…