This extension is required by some programs that depend on openfst that aren't
yet packaged.
* gnu/packages/machine-learning.scm (openfst): Update to 1.8.2.
[arguments]: New field.
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
The extension doesn't build in old versions of openfst such as 1.7.3 without
manually setting CXXFLAGS.
This package definition, including this workaround to build extensions
successfully, will need to be removed once kaldi is fixed upstream and can be
used with up-to-date versions of openfst.
* gnu/packages/machine-learning.scm (openfst-1.7.3)[arguments]: New field.
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
Either kaldi's tests are broken or openblas has reintroduced a bug:
- https://github.com/kaldi-asr/kaldi/pull/4421
More testing & fixes required upstream to re-enable the tests later. For now
they'll just have to be ignored.
* gnu/packages/machine-learning.scm (kaldi): Update to dd107fd.
[arguments]: In 'configure' phase, modify "matrix/Makefile".
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
Kaldi is currently broken with openfst versions newer than 1.7.3 as it changed
some of its function signatures.
Once Kaldi is patched/fixed, this should be removed.
Reference Issues:
- https://github.com/kaldi-asr/kaldi/issues/4131
- https://github.com/kaldi-asr/kaldi/issues/4393
* gnu/packages/machine-learning.scm (openfst-1.7.3): New variable.
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
* gnu/packages/patches/python-pytorch-1.9.0-system-libraries.patch: New file.
* gnu/local.mk (dist_patch_DATA): Add it.
* gnu/packages/machine-learning.scm (python-pytorch-for-r-torch): New variable.
* gnu/packages/machine-learning.scm (python-opentsne): Update to 0.6.1.
[source]<file-name>: Use git-file-name.
[arguments]<#:phases>: Don't return #t. Add 'skip-test phase after 'unpack to
skip a failing test.
* gnu/packages/machine-learning.scm (tensorflow)[arguments]: Rename
'python3.7-compatibility to 'python3.9-compatibility and patch file to allow
building with newer numpy.
This is the result of running:
./pre-inst-env guix style --input-simplification=safe
and manually undoing changes on a dozen of packages to reduce
rebuilds (derivations for emacs, icecat, and libreoffice are unchanged.)
* gnu/packages/machine-learning.scm (python2-scikit-learn)[arguments]: Do not
inherit; remove obsolete 'make-files-writable phase; disable one more failing
test in 'check phase.
* gnu/packages/machine-learning.scm (libsvm)[arguments]: Add build phase
'build-lib; also install header file and shared library in the 'install phase.
There are still many parts of the build that happen sequentially, but at least
this hastens the long build somewhat.
* gnu/packages/machine-learning.scm (tensorflow)[phases]: Remove trailing #t.
{build-pip-package}: Honor the PARALLEL-BUILD? argument.
But keep the older variant for Tensorflow.
* gnu/packages/cpp.scm (abseil-cpp-20200923.3): Renamed older variant.
(abseil-cpp): New variable.
* gnu/packages/machine-learning.scm (tensorflow-lite)
[inputs]: Adjust to use the older abseil-cpp-20200923.3.
* gnu/packages/rpc.scm (grpc-1.16.1)[native-inputs]: Likewise.
Co-authored-by: Greg Hogan <code@greghogan.com>
* gnu/packages/machine-learning.scm (xgboost): New variable.
* gnu/packages/patches/xgboost-use-system-dmlc-core.patch: New patch.
* gnu/local.mk (dist_patch_DATA): Add it.
Signed-off-by: Leo Famulari <leo@famulari.name>
