-
Notifications
You must be signed in to change notification settings - Fork 358
Exclude 'step' from Sensitivity analysis by fixing step at full-progression 'step' #4777
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
|
@sunnyshen321 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D90819468. |
…ession 'step' (facebook#4777) Summary: This diff excludes the "step" feature (MAP_KEY) from sensitivity analysis plots by adding `exclude_map_key` and `target_step` parameters throughout the sensitivity computation. When `exclude_map_key=True` (the new default), the `ax_parameter_sens()` function fixes the step feature at the specified `target_step` value (or the maximum step if not provided) using a `FixedFeatureModel` wrapper, then computes Sobol indices over the remaining parameters. This ensures "step" never appears in the output, making sensitivity plots more interpretable for users who care about parameter effects at final performance rather than across all training steps. Differential Revision: D90819468
8766a45 to
d4aa24a
Compare
…ession 'step' (facebook#4777) Summary: This diff excludes the "step" feature (MAP_KEY) from sensitivity analysis plots by adding `exclude_map_key` and `target_step` parameters throughout the sensitivity computation. When `exclude_map_key=True` (the new default), the `ax_parameter_sens()` function fixes the step feature at the specified `target_step` value (or the maximum step if not provided) using a `FixedFeatureModel` wrapper, then computes Sobol indices over the remaining parameters. This ensures "step" never appears in the output, making sensitivity plots more interpretable for users who care about parameter effects at final performance rather than across all training steps. Differential Revision: D90819468
d4aa24a to
62bcf8f
Compare
…ession 'step' (facebook#4777) Summary: This diff excludes the "step" feature (MAP_KEY) from sensitivity analysis plots by adding `exclude_map_key` and `target_step` parameters throughout the sensitivity computation. When `exclude_map_key=True` (the new default), the `ax_parameter_sens()` function fixes the step feature at the specified `target_step` value (or the maximum step if not provided) using a `FixedFeatureModel` wrapper, then computes Sobol indices over the remaining parameters. This ensures "step" never appears in the output, making sensitivity plots more interpretable for users who care about parameter effects at final performance rather than across all training steps. Differential Revision: D90819468
62bcf8f to
29b8e17
Compare
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #4777 +/- ##
=======================================
Coverage 96.72% 96.72%
=======================================
Files 587 588 +1
Lines 61419 61516 +97
=======================================
+ Hits 59406 59501 +95
- Misses 2013 2015 +2 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
29b8e17 to
0be416e
Compare
…ession 'step' (facebook#4777) Summary: Followup diff from D89385521 This diff excludes the "step" feature (MAP_KEY) from sensitivity analysis plots by adding `exclude_map_key` and `target_step` parameters throughout the sensitivity computation. When `exclude_map_key=True` (the new default), the `ax_parameter_sens()` function fixes the step feature at the specified `target_step` value (or the maximum step if not provided) using a `FixedFeatureModel` wrapper, then computes Sobol indices over the remaining parameters. This ensures "step" doesn't appear in the output by default, making sensitivity plots more interpretable for users who care about parameter effects at final performance rather than across all training steps (should be the most cases). But it also leaves the flexibility for users to set `exclude_map_key=False` if they're interested in the importance of 'step' for some reasons Differential Revision: D90819468
…ession 'step' (facebook#4777) Summary: Followup diff from D89385521 This diff excludes the "step" feature (MAP_KEY) from sensitivity analysis plots by adding `exclude_map_key` and `target_step` parameters throughout the sensitivity computation. When `exclude_map_key=True` (the new default), the `ax_parameter_sens()` function fixes the step feature at the specified `target_step` value (or the maximum step if not provided) using a `FixedFeatureModel` wrapper, then computes Sobol indices over the remaining parameters. This ensures "step" doesn't appear in the output by default, making sensitivity plots more interpretable for users who care about parameter effects at final performance rather than across all training steps (should be the most cases). But it also leaves the flexibility for users to set `exclude_map_key=False` if they're interested in the importance of 'step' for some reasons Differential Revision: D90819468
b725ba3 to
991c9e6
Compare
…ession 'step' (facebook#4777) Summary: Followup diff from D89385521 This diff excludes the "step" feature (MAP_KEY) from sensitivity analysis plots by adding `exclude_map_key` and `target_step` parameters throughout the sensitivity computation. When `exclude_map_key=True` (the new default), the `ax_parameter_sens()` function fixes the step feature at the specified `target_step` value (or the maximum step if not provided) using a `FixedFeatureModel` wrapper, then computes Sobol indices over the remaining parameters. This ensures "step" doesn't appear in the output by default, making sensitivity plots more interpretable for users who care about parameter effects at final performance rather than across all training steps (should be the most cases). But it also leaves the flexibility for users to set `exclude_map_key=False` if they're interested in the importance of 'step' for some reasons Reviewed By: saitcakmak Differential Revision: D90819468
…ession 'step' (facebook#4777) Summary: Followup diff from D89385521 This diff excludes the "step" feature (MAP_KEY) from sensitivity analysis plots by adding `exclude_map_key` and `target_step` parameters throughout the sensitivity computation. When `exclude_map_key=True` (the new default), the `ax_parameter_sens()` function fixes the step feature at the specified `target_step` value (or the maximum step if not provided) using a `FixedFeatureModel` wrapper, then computes Sobol indices over the remaining parameters. This ensures "step" doesn't appear in the output by default, making sensitivity plots more interpretable for users who care about parameter effects at final performance rather than across all training steps (should be the most cases). But it also leaves the flexibility for users to set `exclude_map_key=False` if they're interested in the importance of 'step' for some reasons Reviewed By: saitcakmak Differential Revision: D90819468
991c9e6 to
bae8e13
Compare
…ession 'step' (facebook#4777) Summary: Followup diff from D89385521 This diff excludes the "step" feature (MAP_KEY) from sensitivity analysis plots by adding `exclude_map_key` and `target_step` parameters throughout the sensitivity computation. When `exclude_map_key=True` (the new default), the `ax_parameter_sens()` function fixes the step feature at the specified `target_step` value (or the maximum step if not provided) using a `FixedFeatureModel` wrapper, then computes Sobol indices over the remaining parameters. This ensures "step" doesn't appear in the output by default, making sensitivity plots more interpretable for users who care about parameter effects at final performance rather than across all training steps (should be the most cases). But it also leaves the flexibility for users to set `exclude_map_key=False` if they're interested in the importance of 'step' for some reasons Reviewed By: saitcakmak Differential Revision: D90819468
bae8e13 to
74ed8d2
Compare
|
This pull request has been merged in 4544285. |
Summary:
This diff excludes the "step" feature (MAP_KEY) from sensitivity analysis plots by adding
exclude_map_keyandtarget_stepparameters throughout the sensitivity computation. Whenexclude_map_key=True(the new default), theax_parameter_sens()function fixes the step feature at the specifiedtarget_stepvalue (or the maximum step if not provided) using aFixedFeatureModelwrapper, then computes Sobol indices over the remaining parameters.This ensures "step" never appears in the output, making sensitivity plots more interpretable for users who care about parameter effects at final performance rather than across all training steps.
Differential Revision: D90819468