Fix Multi-Objective cache #872
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
timruhkopf
reviewed
Jul 7, 2022
timruhkopf
reviewed
Jul 7, 2022
timruhkopf
reviewed
Jul 7, 2022
timruhkopf
reviewed
Jul 7, 2022
timruhkopf
reviewed
Jul 7, 2022
timruhkopf
reviewed
Jul 7, 2022
timruhkopf
reviewed
Jul 7, 2022
timruhkopf
reviewed
Jul 7, 2022
| """ | ||
| Transform a multi-objective loss to a single loss. | ||
|
|
||
| Parameters | ||
| ---------- | ||
| values (np.ndarray): Normalized values. | ||
| values : list[float] |
There was a problem hiding this comment.
I would argue, that the normalization is part of the aggregation strategy and should probably be moved here instead
There was a problem hiding this comment.
I don't see a reason why the values should not be normalized here.
timruhkopf
reviewed
Jul 7, 2022
timruhkopf
reviewed
Jul 7, 2022
There was a problem hiding this comment.
Normalization returns a one vector if budget = None. This may not be intuitive and desirable.
-
we might want to make normalization part of the aggregation strategy instead (making it explicit)
-
We could as well call the aggregationStrategy ScalarizationStrategy, implying that we want to make a scalar objective from the multi objective
mfeurer
reviewed
Jul 7, 2022
|
|
||
| # Normalize st all theta values sum up to 1 | ||
| theta = theta / (np.sum(theta) + 1e-10) | ||
|
|
||
| # Weight the values | ||
| theta_f = theta * values | ||
|
|
||
| return np.max(theta_f, axis=1) + self.rho * np.sum(theta_f, axis=1) | ||
| return np.max(theta_f, axis=0) + self.rho * np.sum(theta_f, axis=0) |
There was a problem hiding this comment.
I'm somewhat surprised to see that the summation over an axis changes without a respective test changing. Does this mean that there is no test for ParEGO?
| numerator = data - min_value | ||
| normalized_values.append(numerator / denominator) | ||
| cost = p / q | ||
| costs += [cost] |
renesass
added a commit
that referenced
this pull request
Jul 14, 2022
## Features * [BOinG](https://arxiv.org/abs/2111.05834): A two-stage Bayesian optimization approach to allow the optimizer to focus on the most promising regions. * [TurBO](https://arxiv.org/abs/1910.01739): Reimplementaion of TurBO-1 algorithm. * Updated pSMAC: Can pass arbitrary SMAC facades now. Added example and fixed tests. ## Improvements * Enabled caching for multi-objectives (#872). Costs are now normalized in `get_cost` or optionally in `average_cost`/`sum_cost`/`min_cost` to receive a single float value. Therefore, the cached cost values do not need to be updated everytime a new entry to the runhistory was added. ## Interface changes * We changed the location of Gaussian processes and random forests. They are in the folders `epm/gaussian_process` and `epm/random_forest` now. * Also, we restructured the optimizer folder and therefore the location of the acquisition functions and configuration chooser. * Multi-objective functions are located in the folder `multi_objective`. * pSMAC facade was moved to the facade directory. Co-authored-by: Difan Deng <[email protected]> Co-authored-by: Eddie Bergman <[email protected]> Co-authored-by: Carolin Benjamins <[email protected]> Co-authored-by: timruhkopf <[email protected]>
sharpe5
added a commit
to sharpe5/SMAC3
that referenced
this pull request
Aug 11, 2022
Version 1.4.0 (automl#730, automl#855, automl#869, automl#872)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Closes #852.
I slightly rearranged the runhistory so that it's more intuitive.
What else did I do?
average_cost/min_cost/sum_costreturn a list of floats (instead of a float) now.For example, for
average_cost, each objective is averaged separately. Imagine you have two runs with [100, 200] and [0, 0] then you'd get [50, 100]. However, once you callget_cost, this (cached) [50, 100] is normalized based on all passed entries. That means thatget_costreturns a float in the end.Edit: I added an argument in
average_cost/min_cost/sum_costto return the normalized (single float) values. Makes the code easier and scalable.One more thing which really is weird: In the runhistory2epm we iterate over the run dictionary (runkey, runvalue). But here we use the
run.costdirectly and hence we have to callnormalize_costsin the case of MO again. I wonder why this is done this way? @mfeurer