Ecosyste.ms: Issues
An open API service for providing issue and pull request metadata for open source projects.
GitHub / f-dangel/curvlinops issues and pull requests
#125 - [FIX] Output shapes of KFAC's `matmat`/`matvec`
Pull Request -
State: closed - Opened by runame about 1 month ago
- 1 comment
Labels: bug
#124 - `A.torch_matvec()` changes the shape of tensors
Issue -
State: closed - Opened by BrunoKM about 1 month ago
- 5 comments
Labels: bug
#123 - [REF] Migrate from `setup.py` to `pyproject.toml` (fixes #122)
Pull Request -
State: closed - Opened by kourbou about 2 months ago
- 9 comments
#122 - Error when installing `curvlinops` as dependency
Issue -
State: closed - Opened by wiseodd 3 months ago
- 4 comments
#121 - [REF] Deprecate `pkg_resources`, use `importlib` & `packaging`
Pull Request -
State: closed - Opened by f-dangel 3 months ago
- 10 comments
#120 - [REF] Improve numerical stability of Frobenius estimator
Pull Request -
State: closed - Opened by f-dangel 3 months ago
- 1 comment
#119 - [FIX] Checks if `Enum` contains string
Pull Request -
State: closed - Opened by runame 3 months ago
- 7 comments
Labels: bug
#118 - [BUG] Check if string is in `Enum` incompatible with older Python versions
Issue -
State: closed - Opened by runame 3 months ago
Labels: bug
#117 - Remove `loss_average` argument of `KFACLinearOperator`
Pull Request -
State: closed - Opened by runame 4 months ago
- 1 comment
Labels: enhancement
#116 - Implement EKFAC
Issue -
State: open - Opened by runame 4 months ago
Labels: enhancement
#115 - [ADD] Implement adjoint for sub-matrix linear operator
Pull Request -
State: closed - Opened by f-dangel 4 months ago
- 1 comment
#114 - [ADD] `state_dict` functionality to `KFACLinearOperator` and `KFACInverseLinearOperator`
Pull Request -
State: closed - Opened by runame 4 months ago
- 1 comment
Labels: enhancement
#113 - Add `state_dict` functionality to `KFACLinearOperator` and `KFACInverseLinearOperator`
Issue -
State: closed - Opened by runame 4 months ago
Labels: enhancement
#112 - [FIX] Scaling of MC Fisher for `MSELoss` and `BCEWithLogitsLoss` with mean reduction
Pull Request -
State: closed - Opened by runame 4 months ago
- 1 comment
Labels: bug
#111 - [BUG] `FisherMCLinearOperator` is off by a factor of `C` when using `MSELoss` or `BCEWithLogitsLoss` with mean reduction
Issue -
State: closed - Opened by runame 4 months ago
Labels: bug
#110 - [FIX] KFAC scale for `loss_average="batch+sequence"`
Pull Request -
State: closed - Opened by runame 5 months ago
- 1 comment
Labels: bug
#109 - [FIX] Scaling of empirical/MC Fisher for output with more than two dimensions
Pull Request -
State: closed - Opened by runame 5 months ago
- 1 comment
Labels: bug
#108 - [BUG] Scaling of (KFAC) empirical/MC Fisher broken in some cases when using mean reduction for loss
Issue -
State: closed - Opened by runame 5 months ago
Labels: bug
#107 - [BUG] Scaling of gradient covariance Kronecker factor in KFAC for `loss_average="batch+sequence"` can be off
Issue -
State: closed - Opened by runame 5 months ago
Labels: bug
#106 - [ADD] `LSMRInverseLinearOperator`
Pull Request -
State: closed - Opened by runame 5 months ago
- 1 comment
Labels: enhancement
#105 - [ADD] `LSMRInverseLinearOperator`
Issue -
State: closed - Opened by runame 5 months ago
Labels: enhancement
#104 - [REF] Make KFAC computation side effect free
Pull Request -
State: closed - Opened by f-dangel 5 months ago
- 3 comments
#103 - [FIX] Device errors in tests
Pull Request -
State: closed - Opened by f-dangel 5 months ago
- 1 comment
#102 - [ADD] Option to retry failed Cholesky decompositions in double precision in `KFACInverseLinearOperator`
Pull Request -
State: closed - Opened by runame 5 months ago
- 1 comment
Labels: enhancement
#101 - Numerical stability of `cholesky` in `KFACInverseLinearOperator`
Issue -
State: closed - Opened by runame 5 months ago
Labels: enhancement
#100 - Support arbitrary `X` in `data`
Pull Request -
State: closed - Opened by wiseodd 6 months ago
- 9 comments
Labels: enhancement
#99 - [ADD] Support `BCEWithLogitsLoss` in Fisher-MC and KFAC
Pull Request -
State: closed - Opened by f-dangel 6 months ago
- 1 comment
Labels: enhancement
#98 - Add `fisher_type="forward-only"` (FOOF method) option to `KFACLinearOperator`
Pull Request -
State: closed - Opened by runame 6 months ago
- 2 comments
Labels: enhancement
#97 - [CI] Run GH actions for PR to any branch
Pull Request -
State: closed - Opened by f-dangel 6 months ago
- 1 comment
#96 - Add `trace`, `det`, `logdet`, and `frobenius_norm` properties to `KFACLinearOperator`
Issue -
State: closed - Opened by runame 6 months ago
Labels: enhancement
#95 - Add `trace`, `det`, `logdet`, and `frobenius_norm` property to `KFACLinearOperator`
Pull Request -
State: closed - Opened by runame 6 months ago
- 2 comments
Labels: enhancement
#94 - Potentially switch to or add option to use `cholesky_solve` instead of `cholesky_inverse` in `KFACInverseLinearOperator`
Issue -
State: open - Opened by runame 6 months ago
Labels: enhancement
#93 - Add heuristic and exact damping to `KFACInverseLinearOperator`
Pull Request -
State: closed - Opened by runame 6 months ago
- 1 comment
Labels: enhancement
#92 - Add option for heuristic and exact damping to `KFACInverseLinearOperator`
Issue -
State: closed - Opened by runame 6 months ago
Labels: enhancement
#91 - Allow for matrix-matrix and matrix-vector products with `KFACLinearOperator` and `KFACInverseLinearOperator` without converting to numpy
Pull Request -
State: closed - Opened by runame 6 months ago
- 1 comment
Labels: enhancement
#90 - Support `BCEWithLogitsLoss`
Issue -
State: closed - Opened by f-dangel 6 months ago
Labels: enhancement
#89 - [REF] Parallelize `sample_grad_output` in MC Fisher
Pull Request -
State: closed - Opened by f-dangel 6 months ago
#88 - [REF] Parallelize `grad_output` computation in empirical Fisher
Pull Request -
State: closed - Opened by f-dangel 6 months ago
- 1 comment
#87 - [OPT] Remove `for` loop to compute `grad_outputs` in EF
Issue -
State: closed - Opened by f-dangel 6 months ago
Labels: enhancement
#86 - Make less/no assumption about the data
Issue -
State: closed - Opened by wiseodd 6 months ago
- 3 comments
Labels: enhancement
#85 - [REF] Remove for-loop over samples in MC Fisher
Pull Request -
State: closed - Opened by f-dangel 7 months ago
Labels: enhancement
#84 - [REF] Improve efficiency of empirical Fisher
Pull Request -
State: closed - Opened by f-dangel 7 months ago
- 1 comment
Labels: enhancement
#83 - Improve efficiency of MC Fisher
Issue -
State: closed - Opened by f-dangel 7 months ago
Labels: enhancement
#82 - KFAC factors are twice as large as BackPACK's and ASDL's
Issue -
State: closed - Opened by wiseodd 7 months ago
- 1 comment
#81 - [DOC] Add links to source code
Pull Request -
State: closed - Opened by f-dangel 7 months ago
- 1 comment
#80 - [ADD] Squared Frobenius norm estimator
Pull Request -
State: closed - Opened by f-dangel 7 months ago
- 3 comments
#79 - [REF | KFAC] Use module names instead of `data_ptr()`s for mapping
Pull Request -
State: closed - Opened by f-dangel 7 months ago
#78 - [FIX] Bug #77: Invalidated parameter mapping after device transfer in KFAC
Pull Request -
State: closed - Opened by f-dangel 7 months ago
- 2 comments
#77 - [BUG | KFAC] Changing device invalidates parameter mapping
Issue -
State: closed - Opened by f-dangel 7 months ago
Labels: bug
#76 - [FIX] `Generator` device for sampling
Pull Request -
State: closed - Opened by runame 7 months ago
Labels: bug
#75 - Verify that the order of the mini-batches is deterministic where it matters
Issue -
State: open - Opened by runame 7 months ago
Labels: enhancement
#74 - [ADD] Support block-diagonal Hessian approximation
Pull Request -
State: closed - Opened by f-dangel 7 months ago
- 1 comment
#73 - [REF] Implement `_matmat` instead of `_matvec`
Pull Request -
State: closed - Opened by f-dangel 8 months ago
- 3 comments
#72 - [FMT] Apply `black==24.1.1`
Pull Request -
State: closed - Opened by f-dangel 8 months ago
- 1 comment
#71 - Unnecessary device transfer in `_postprocess` method
Issue -
State: open - Opened by runame 8 months ago
- 3 comments
Labels: enhancement
#70 - [ADD] Option to specify number of data points
Pull Request -
State: closed - Opened by f-dangel 8 months ago
- 1 comment
#69 - [ADD] `KFACInverseLinearOperator`
Pull Request -
State: closed - Opened by runame 8 months ago
- 1 comment
Labels: enhancement
#68 - Add `KFACInverseLinearOperator` class
Issue -
State: closed - Opened by runame 8 months ago
Labels: enhancement
#67 - Update `main` with `development` in preparation of `1.2.0` release
Pull Request -
State: closed - Opened by f-dangel 8 months ago
#66 - [DOC] Update changelog & contributors, fix PyPI upload, prepare `1.2.0`
Pull Request -
State: closed - Opened by f-dangel 8 months ago
- 1 comment
#65 - [ADD] Linear operator for Hessian w.r.t. intermediate features
Pull Request -
State: closed - Opened by f-dangel 8 months ago
- 1 comment
#64 - Add support for KFAC for `Conv2d` modules
Pull Request -
State: closed - Opened by runame 8 months ago
- 5 comments
Labels: enhancement
#63 - Add support for KFAC-expand and KFAC-reduce for `Linear` modules
Pull Request -
State: closed - Opened by runame 8 months ago
- 3 comments
Labels: enhancement
#62 - Support KFAC for >2d model outputs
Pull Request -
State: closed - Opened by runame 8 months ago
- 1 comment
Labels: enhancement
#61 - [FIX] Typo
Pull Request -
State: closed - Opened by runame 11 months ago
#60 - Support >2d model outputs for KFAC with type-2 Fisher
Issue -
State: closed - Opened by runame 11 months ago
Labels: enhancement
#59 - [ADD] Support KFAC if the network has inplace activations
Pull Request -
State: closed - Opened by f-dangel 11 months ago
- 3 comments
#58 - Bug: Networks with in-place operations not supported in KFAC
Issue -
State: closed - Opened by f-dangel 11 months ago
Labels: bug
#57 - [ADD] Option to treat weight and bias of a layer jointly in KFAC
Pull Request -
State: closed - Opened by f-dangel 11 months ago
- 2 comments
#56 - [ADD] Support for KFAC with type-2 Fisher
Pull Request -
State: closed - Opened by runame 11 months ago
- 2 comments
Labels: enhancement
#55 - [ADD] Allow subsets of parameters (weight-only, bias-only) in KFAC
Pull Request -
State: closed - Opened by f-dangel 11 months ago
- 1 comment
#54 - [ADD] Support for empirical Fisher in KFAC
Pull Request -
State: closed - Opened by runame 11 months ago
- 1 comment
Labels: enhancement
#53 - Support treating weights and biases of a layer jointly in KFAC
Issue -
State: closed - Opened by f-dangel 11 months ago
Labels: enhancement
#52 - [ADD] Support `CrossEntropyLoss` in KFAC
Pull Request -
State: closed - Opened by f-dangel 11 months ago
- 1 comment
#51 - [ADD] Support arbitrarily-ordered `params` in KFAC
Pull Request -
State: closed - Opened by f-dangel 11 months ago
#50 - Support KFAC-expand and KFAC-reduce for linear weight-sharing layers.
Issue -
State: closed - Opened by runame 11 months ago
Labels: enhancement
#49 - Support type-2 Fisher in KFAC
Issue -
State: closed - Opened by runame 11 months ago
- 5 comments
Labels: enhancement
#48 - Support linear layers without bias or weights in KFAC
Issue -
State: closed - Opened by f-dangel 11 months ago
Labels: good first issue
#47 - Allow arbitrarily-ordered parameters in KFAC
Issue -
State: closed - Opened by f-dangel 11 months ago
Labels: good first issue
#46 - Support using the empirical distribution for gradient covariances in KFAC
Issue -
State: closed - Opened by f-dangel 11 months ago
Labels: good first issue
#45 - Support KFC (Grosse 2016) for convolutional layers in KFAC
Issue -
State: closed - Opened by f-dangel 11 months ago
#44 - Support `nn.CrossEntropyLoss` in KFAC
Issue -
State: closed - Opened by f-dangel 11 months ago
#43 - [ADD] Minimal prototype for KFAC
Pull Request -
State: closed - Opened by f-dangel 11 months ago
- 1 comment
#42 - Linear operator for KFAC
Issue -
State: closed - Opened by f-dangel 11 months ago
Labels: enhancement
#41 - Feature request: add method to convert to matrix
Issue -
State: closed - Opened by andres-fr 11 months ago
- 4 comments
#40 - [ADD] Hutchinson-style matrix diagonal estimation
Pull Request -
State: closed - Opened by f-dangel 11 months ago
- 1 comment
#39 - [ADD] Implement Hutch++ trace estimation
Pull Request -
State: closed - Opened by f-dangel 11 months ago
- 1 comment
#38 - [ADD] Vanilla Hutchinson trace estimation
Pull Request -
State: closed - Opened by f-dangel 11 months ago
- 1 comment
#37 - [DOC] Example on Fisher-averaging
Pull Request -
State: closed - Opened by f-dangel 11 months ago
- 1 comment
#36 - [REF] Use built in functorch, use BackPACK 1.6
Pull Request -
State: closed - Opened by f-dangel about 1 year ago
- 1 comment
#35 - [ADD] Implement `_adjoint` for Jacobians
Pull Request -
State: closed - Opened by f-dangel about 1 year ago
#34 - [ADD] Linear operator for transposed output-parameter Jacobian of a NN
Pull Request -
State: closed - Opened by f-dangel about 1 year ago
- 1 comment
#33 - [ADD] Implement and test `_adjoint`
Pull Request -
State: closed - Opened by f-dangel about 1 year ago
- 1 comment
#32 - [ADD] Output-parameter Jacobian of an NN, deprecate python 3.7
Pull Request -
State: closed - Opened by f-dangel about 1 year ago
- 1 comment
#31 - Feature request: Diagonal estimation algorithms
Issue -
State: closed - Opened by f-dangel about 1 year ago
- 2 comments
Labels: good first issue
#30 - Feature request: Trace estimation algorithms
Issue -
State: closed - Opened by f-dangel about 1 year ago
- 7 comments
#29 - [DOC] Add Neumann inverse to natural gradient example
Pull Request -
State: closed - Opened by f-dangel over 1 year ago
- 1 comment
#28 - [ADD] Linear operator for Neumann inverse
Pull Request -
State: closed - Opened by f-dangel over 1 year ago
- 2 comments
#27 - [ADD] Partial boundary specification
Pull Request -
State: closed - Opened by f-dangel over 1 year ago
- 1 comment
#26 - [DOC] Add sub-matrix example
Pull Request -
State: closed - Opened by f-dangel over 1 year ago
- 1 comment