-
Notifications
You must be signed in to change notification settings - Fork 479
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
add test loss #1073
add test loss #1073
Conversation
This pull request was exported from Phabricator. Differential Revision: D41202866 |
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Reviewed By: cyrjano Differential Revision: D41202866 fbshipit-source-id: 48c661847b0340163bdcff69a0643df84e7c6d11
ee140d2
to
f316335
Compare
This pull request was exported from Phabricator. Differential Revision: D41202866 |
1 similar comment
This pull request was exported from Phabricator. Differential Revision: D41202866 |
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Reviewed By: cyrjano Differential Revision: D41202866 fbshipit-source-id: 97713fac4ece5e8f2192ea7cddb7d05eb3cc3e06
f316335
to
7796dc1
Compare
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Differential Revision: https://internalfb.com/D41202866 fbshipit-source-id: 68898a2a57f9a3599a06902c4eeb9d7c748f53f8
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Differential Revision: https://internalfb.com/D41202866 fbshipit-source-id: 3e085383ef6a735695a190d4289dd4c702702f06
This pull request was exported from Phabricator. Differential Revision: D41202866 |
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Reviewed By: cyrjano Differential Revision: D41202866 fbshipit-source-id: 292f72b7f217b2ee111bb29536551206db7cc6fe
7796dc1
to
c642664
Compare
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Differential Revision: https://internalfb.com/D41202866 fbshipit-source-id: a7a82afa2cae0b8cad036980a7a823aeb9d68ab7
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Differential Revision: https://internalfb.com/D41202866 fbshipit-source-id: 3b3dbafe332d4abbcde9eb08c2ae193262104848
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Differential Revision: https://internalfb.com/D41202866 fbshipit-source-id: 53c6e58fadc9b3cce4f5e931f47edd0f2f3fd718
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Differential Revision: https://internalfb.com/D41202866 fbshipit-source-id: 80ede21e481d4052b278e6ef3be4d8a19959bcd3
This pull request was exported from Phabricator. Differential Revision: D41202866 |
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Reviewed By: cyrjano Differential Revision: D41202866 fbshipit-source-id: ada2fa1fec7d2414be069797f676097bac0a1cf6
c642664
to
2854b54
Compare
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Differential Revision: https://internalfb.com/D41202866 fbshipit-source-id: 84827a117785e17b6dabb8b9b3acb94f1aa686af
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Differential Revision: https://internalfb.com/D41202866 fbshipit-source-id: e6a261797c7e89d03e40b026001e00d7ec30853e
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Differential Revision: https://internalfb.com/D41202866 fbshipit-source-id: 588dc003e9e5685ab190667f56a7447cd4509632
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Reviewed By: cyrjano Differential Revision: D41202866 fbshipit-source-id: 4258e3597b1f2e30ba5059bb7d440c8de7fd3ac1
This pull request was exported from Phabricator. Differential Revision: D41202866 |
2854b54
to
eb50036
Compare
Summary: Pull Request resolved: pytorch#1073 - For all `TracInCPBase` implementations, this adds an additional `test_loss_fn` initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient of `loss_fn` for a given training example, multiplied with the gradient of `test_loss_fn` for a given test example. Before, `test_loss_fn` was assumed to be the same as `loss_fn`. - checks regarding the reduction type of both `loss_fn` and `test_loss_fn` are now handled by helper functions `_check_tracincp_loss_fn` and `_check_tracincp_fast_loss_fn`. - documentation is updated. one detail: for `TracInCP`, we assume that `sample_wise_grads_per_batch` is applied to both `loss_fn` and `test_loss_fn` (if provided), and this is mentioned in the documentation. - `test_tracin_regression.test_tracin_regression` is slightly modified - `DataInfluenceConstructor` now can explicitly pass in the same loss function for both `loss_fn` and `test_loss_fn` (done when `duplicate_loss_fn=True`). Doing so would have the same effect as not passing in `test_loss_fn`, so the original tests are also applied to the case when `duplicate_loss_fn=True`, as the expected behavior should be the same as before. - a new test, `test_tracin_regression.test_tracin_constant_test_loss_fn` is added. For all implementations of `TracInCPBase`, it checks that if `test_loss_fn` is a constant loss function, the influence scores are all 0's. This should be the case, because if `test_loss_fn` is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples. Differential Revision: https://internalfb.com/D41202866 fbshipit-source-id: e9eaf630be6b54f7d0c86c681dcf0c9e6dddb6f1
This pull request has been merged in 508ee12. |
Summary:
TracInCPBase
implementations, this adds an additionaltest_loss_fn
initialization argument, which is the loss function to apply to test examples when computing the influence of a training example on a test example. With this change,the influence score is a sum over terms for each checkpoint, where each term is the gradient ofloss_fn
for a given training example, multiplied with the gradient oftest_loss_fn
for a given test example. Before,test_loss_fn
was assumed to be the same asloss_fn
.loss_fn
andtest_loss_fn
are now handled by helper functions_check_tracincp_loss_fn
and_check_tracincp_fast_loss_fn
.TracInCP
, we assume thatsample_wise_grads_per_batch
is applied to bothloss_fn
andtest_loss_fn
(if provided), and this is mentioned in the documentation.test_tracin_regression.test_tracin_regression
is slightly modified -DataInfluenceConstructor
now can explicitly pass in the same loss function for bothloss_fn
andtest_loss_fn
(done whenduplicate_loss_fn=True
). Doing so would have the same effect as not passing intest_loss_fn
, so the original tests are also applied to the case whenduplicate_loss_fn=True
, as the expected behavior should be the same as before.test_tracin_regression.test_tracin_constant_test_loss_fn
is added. For all implementations ofTracInCPBase
, it checks that iftest_loss_fn
is a constant loss function, the influence scores are all 0's. This should be the case, because iftest_loss_fn
is constant, its gradients would all be 0's, so that training examples have 0 influence on test examples.Reviewed By: cyrjano
Differential Revision: D41202866