Add a simple getting started notebook #129
Conversation
There was a problem hiding this comment.
Copilot wasn't able to review any files in this pull request.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
am-kaiser
left a comment
am-kaiser
left a comment
There was a problem hiding this comment.
Very good job. I have a couple of comments but many of them are just small changes.
| # 1. Runs the simulator `N_SIMULATIONS_PER_POINT` times at each training point `x` | ||
| # 2. Fits the Gumbel distribution to those samples to estimate `loc(x)` and `scale(x)` | ||
| # 3. Provides these fitted parameters to the GP | ||
| # |
There was a problem hiding this comment.
make_experiment does not actually run anything. It is just a wrapper to simplify the steps you mention in later stages
There was a problem hiding this comment.
Is it better now?
| # 4. Extracts the **maximum** response in each period | ||
| # 5. Estimates the **median** of the resulting extreme response distribution | ||
| # 6. Repeats under different draws from the GP posterior to capture model uncertainty | ||
| # |
There was a problem hiding this comment.
This sounds a bit like MarginalCDFExtrapolation is basically brute force just with the surrogate model. That is not quite correct (GPBruteForce is doing that more or less). The procedure for MarginalCDFExtrapolation is:
- from the GP get the posterior distributions at the specified env. points
- draw samples from posterior distributions
- define Gumbel distribution using loc and scale
- using inverse CDF we can then calculate QoI for each posterior sample
This is just an explanation for you but maybe a bit too detailed for such a tutorial. While the batching is done and has some effects. It is maybe not super important at this level of abstraction
There was a problem hiding this comment.
Thank you for the explanation. I tried to simplify the text for this tutorial.
|
I have addressed all the review comments. Please take another look! Thanks! |
Add getting started notebook for Axtreme (a simplification of the basic example).