[scikit-learn] annotated tag debian/0.6.0.dfsg-1 created (now 2a2ee13)
Andreas Tille
tille at debian.org
Wed Dec 28 13:11:15 UTC 2016
This is an automated email from the git hooks/post-receive script.
tille pushed a change to annotated tag debian/0.6.0.dfsg-1
in repository scikit-learn.
at 2a2ee13 (tag)
tagging 03a85c19ac2854f2a33f613f87e81fd5f4560f55 (commit)
replaces debian/0.5-1
tagged by Yaroslav Halchenko
on Mon Jan 10 09:22:35 2011 -0500
- Log -----------------------------------------------------------------
Debian release 0.6.0.dfsg-1
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
iEYEABECAAYFAk0rFisACgkQjRFFY3XAJMhwoQCeIEcXo2KA4zv2ie4T+K1rAusf
NmgAni+Es6loKf7LInNu6XciBOeA5k/8
=fMhh
-----END PGP SIGNATURE-----
Alexandre Gramfort (69):
FIX: ticket 147 on pb with 2d y in f_regression
FIX: ticket 147 on pb with 2d y in f_classif
adding 'iid' option in cross_val_score
Merge branch 'master' of git at github.com:scikit-learn/scikit-learn
pretifying plot_weighted_classes.py
FIX: quick fix in predict_proba in LogisticRegression
removing debug compile flags
sgd module code review
sgd module code review
adding path example on logistic on IRIS dataset
Merge branch 'sgd'
increasing precision in plot_logistic_path.py to get nicer path
DOC: spelling
ENH : pyflakes on examples to avoid useless imports + addint print __doc__
ENH : more love in examples (adding print __doc__ + some brief descriptions in headers + fixing Anova SVC Pipeline example)
FIX: fix docstrings in LARS (issue 8 on github)
cosmit + typos in doc
Merge branch 'master' of github.com:scikit-learn/scikit-learn
ENH: adding partial support for predict_log_proba in Pipeline and log reg
rewriting f_oneway in the scikit to avoid useless recomputations
Merge branch 'master' into log_proba
adding comment to explain the reimplementation of f_oneway
Merge branch 'master' of github.com:scikit-learn/scikit-learn
Merge branch 'master' into log_proba
typo
removing use_svd option in LDA. Only scipy SVD is supported.
Merge branch 'master' into log_proba
ENH: adding predict_log_proba to LDA and QDA + tests to reach 100% coverage
ENH : adding support for predict_log_proba in Naive Bayes
ENH: adding support for predict_log_proba in SVC and sparse.SVC
ENH : adding predict_log_proba in sparse logistic regression
FIX: make sure class_weight='auto' do not change the result for balanced problems
Merge branch 'master' into log_proba
API : implement coef_init as fit parameter in glm.coordinate_descent module.
API: exposing fit_intercept params in LassoCV and ElasticNetCV
ENH : adding test in pipeline + increase coverage
fix doc generation pb introduced by previous commit
FIX: fix class weight auto
pep8 in plot_weighted_samples.py
ENH : adding kneighbors_graph to build the graph of neighbors as a sparse matrix
FIX fragile doctest
ENH : adding NeighborsBarycenter for regression pbs using k-Nearest Neighbors
DOC: adding NeighborsBarycenter to doc
DOC: better docstring for barycenter_weights function
DOC: even better docstrings in neighbors
MISC: reindenting BallTree C++ code (no tabs + 4 spaces)
DOC : more on docstrings in neighbors.py
review of gaussian process module
API renmae k->n_neighbors
Merge branch 'log_proba'
ENH : improving the speed of ridge with inplace computation + symmetric pos def constraint
Merge branch 'neighbor_barycenter'
ENH : coordinate descent speed up when n_samples > n_features in cd_fast.pyx
ENH : allowing Gram matrix precomputing in Lasso / ElasticNet to
Merge branch 'master' of github.com:scikit-learn/scikit-learn
ENH : speed improvement in lasso_path with precomputed gram matrix
Merge branch 'master' of github.com:scikit-learn/scikit-learn
pep8 in coordinate_descent.py
pep8 + N->n_samples and D->n_features
Merge branch 'master' of github.com:scikit-learn/scikit-learn
giving more love to benchmarks (pep8, pyflakes, var names, etc...)
Merge branch 'master' of github.com:scikit-learn/scikit-learn
revert previous commit regarding mpl_toolkits.mplot3d in bench
API : maxit replaced by max_iter everywhere
ENH : new scikits.learn.metrics.pairwise module
Merge branch 'master' of https://github.com/dubourg/scikit-learn into dubourg-master
pyflakes in plot_gp_diabetes_dataset.py
renaming plot_gp_diabetes_dataset.py as nothing is plotted
FIX : fix extra parenthesis in mixture ...
Alexandre Passos (3):
Adding random projections SVD to scikits.learn.pca as an option
Adding the power iteration parameter to fast_svd (to make it better in high-rank very-big very-sparse matrices according to the Martinsson et al survey
Merging the rng changes
Anne-Laure Fouque (3):
ENH added R^2 coeff which is now used as score function in RegressorMixin
renamed explained_variance_score to r2_score in linear_model
adding r2_score : fixed typos and doctest
Bertrand Thirion (9):
cosmit in pca and test_pca
merged origin and fixed a conflict
merged origin
Merge branch 'master' of github.com:scikit-learn/scikit-learn
Merge branch 'master' of github.com:scikit-learn/scikit-learn
Merge branch 'master' of github.com:scikit-learn/scikit-learn
Merge branch 'master' of github.com:scikit-learn/scikit-learn
merged the mainr epo
Merge branch 'master' of github.com:scikit-learn/scikit-learn
Christian Osendorfer (1):
Fixed problem with big full covariance matrices: sum,log instead of log,prod for loglikelihood computations.
Fabian Pedregosa (170):
Start of 0.6 development cycle
Add note on executing the test suite.
Update web page.
Add a note on complexity for SVMs.
Add datasets to __init__ file.
Correct typo in docstring.
Allow access to multi-class SVM in liblinear.
Do not execute test coverage by default.
lighten GMM tests.
remove n_dim property (use plain field).
Fix and enable _covar_mstep_full in gmm.py
Cosmetic changes.
Bindings for libsvm-dense
Update svm benchmark with latests libsvm.
Some fixes for libsvm-dense
More accurate info in examples.
Update svm examples affected by latest API changes.
DOC: Some docstring for libsvm low level API.
Revert "DOC: Some docstring for libsvm low level API."
Revert "Update svm examples affected by latest API changes."
Revert "More accurate info in examples."
Revert "Some fixes for libsvm-dense"
Revert "Update svm benchmark with latests libsvm."
Revert "Bindings for libsvm-dense"
ENH: enhacements in the gmm module.
Make previous commit work also with old versions of scipy.
No specific need that matrix is upper-triangular in gmm.
Fix doctests in gmm (skip random ones).
Revert "Fix doctests in gmm (skip random ones)."
Revert "No specific need that matrix is upper-triangular in gmm."
Revert "Make previous commit work also with old versions of scipy."
Revert "ENH: enhacements in the gmm module."
Bindings for libsvm-dense
Update svm benchmark with latests libsvm.
Some fixes for libsvm-dense
More accurate info in examples.
Update svm examples affected by latest API changes.
DOC: Some docstring for libsvm low level API.
Compile _libsvm_sparse in the sparse module.
Add setup.py to svm.sparse
Preliminary fix for naming issue in OSX with libsvm.
Add a namespace to svm methods to avoid same name mangling.
Fix for building libsvm in a portable way.
FIX: fix doctest with recent API changes.
FIX: fix fragile doctest.
Updated liblinear to latest version 1.7.
Make liblinear quieter.
Update classes to use new features from liblinear 1.7.
Move logistic into glm and add a sparse version.
Doc: better tests for logistic.
Fix imports in example.
Fix doctests in sgd module.
Welcome Peter.
Avoid iterating over features in gmm.
Add more sanity checks for svm with precomputed kernels.
Use n_jobs=1 as default value in SGD module.
Unique URL for release-specific doc
Cleanup in libsvm bindings.
Cosmetic changes in gmm.
Improve docstrings in metrics.py
Cosmetic changes
DOC: Add new installation media and a note for pythonxy users.
FIX: prefix with plot examples that produce output image.
New implementation of LARS algorithm.
Add a test for lars_path.
Fix typo (wantto -> want to)
remove obsolete bench_lars.py
FIX: replace nsamples --> n_samples in svm docstrings.
Remove BaseLib class.
Implement make html-noplot for building the doc.
Update libsvm docstring with latest API changes.
Rename predict_margin --> decision_function.
Indentation fixes in libsvm bindings.
Performance improvements in LARS.
Better heuristic in LARS.
Add support for np.float32 matrices in lars_path.
Add parameter precompute='auto' for *LARS classes.
Some LARS refactoring.
Rename scikits.learn.gmm to scikits.learn.mixture.
Update developers info.
Add GridSearch and GridSearchCV to the class reference.
Update svm docs (content of dual_coef_).
Account for lower=True option in solve_triangular.
Do not import gaussian_process from top level __init__.
update NuSVC docstring.
Fix failing doctests in gaussian_process.rst.
Fix GridSearch does not exist.
Give credit for web page layout.
glm --> linear_model rename holocaust.
Welcome Vincent Dubourg.
Update AUTHORS information.
Initial support for weighted samples in svm module.
Cosmetic changes to web page layout.
Fix example paths for GMM after renaming.
Update class reference list.
Cosmetic changes in documentation.
Add sgd.* to class reference.
Move benchmarks outside the source tree.
Fix precompute keyword in LARS.
Update LARS benchmarks with latest API changes.
Cosmetic changes in plot_weighted_samples.py
Add cross-references between LassoLARS and Lasso.
More rename in the sgd module.
ENH: prettify web page layout.
Some love for scikits.learn.svm.
FIX web page layout for very long paths.
Update LARS documentation.
Fix for linear_model.rst
More love for rst docs.
Like it or not, we depend on setuptools.
Use original diabetes data as shipped by the R package lars.
rename lars --> least_angle
Remove duplicates in linear_model/__init__.py
Use relative imports in datasets.
FIX: sparse svms do not accept callable kernels.
py3k fixes: callable has been removed.
Py3k compatibility
Remove redundant site.cfg parsing.
Update status of py3k support.
Cosmetic changes in LARS.
FIX: correctly add depends files to setup.py.
Make libsvm recognize labels in increasing order.
Correct array size in decision_function docstring
TEST: sanity check on decision_function.
Inverse sign in decision_function.
No need to sort predict_proba any more.
Add a comment on inverting the sign of decision_function.
FIX: order of indices of support vectors in multiclass.
Shuffle globally for iris in test_svm.
Divide parameter alpha / n_samples for consistency with Lasso.
Cosmetic changes.
Cosmetic changes in lars.
Update .mailmap
FIX: fix bug in sparse liblinear: bias parameter was not set.
FIX lda, qda: new numpy.bincount requires integer arguments.
Started Changelog 0.6.
Change link in plot_face_recognition.
Remove example plot_lar.py
FIX: do not invert the sign of decision_function in OneClasSVM.
Add missing options to OneClassSVM.
web page layout fixes.
Remove duplicate docs (sphinx generates this for us).
Prepare for 0.6 release.
Remove generated classes on make clean.
Add notes on fluctiations of liblinear.
Add type info to docstrings.
FIX: backwards compatibility for scipy <= 0.8
Remove Methods from docstring.
FIX: scipy 0.9 compatibility fixes
FIX: second argument in euclidean_distances.
Cosmetic changes.
Better version detection for scipy
FIX: stupid mistake.
FIX Stupid mistake
More robust utils.fixes.
FIX: docstring.
FIX: np.unique.
Change version to 0.6-rc
Add AUTHORS to web page.
Note on LinearSVC.
Web page layout.
FIX: update to latest API.
Web page update.
FIX tests when run with scikits.learn.test()
Update doc.
Do not support precomputed kernels in sparse matrices.
Update version information to 0.6.0
Update Mailmap.
Update authors list.
Update README.
Gael Varoquaux (55):
ENH: Small optimization to BaseEstimator
BUG: Make sure that grid_search works with sparse data.
MISC: Cosmit in new GMM classifier example
DOC: Make the plot_ica_vs_pca example richer.
MISC: Some tweeks to the layout so that the docs display better on a
DOC: Fix title level in install
DOC: make the index page content clearer
MISC: Explicit acronym
MISC: PEP8 in docs
DOC: Change the titles' layout
DOC: Rewamp the tables of contents and corresponding layout
Merge branch 'master' of github.com:scikit-learn/scikit-learn
DOC: Make sure the docstring of pca render well
DOC: Remove empty section
DOC: Add documentation for ICA/PCA
DOC: Remove useless tables of contents
DOC: work on the clustering documentation
Merge branch 'master' of github.com:scikit-learn/scikit-learn
DOC: Tweak in the clustering docs.
DOC: document with more details the GMM module.
Merge branch 'master' of github.com:scikit-learn/scikit-learn
Merge branch 'master' of github.com:scikit-learn/scikit-learn
DOC: make the neighbors doc sexier
Merge branch 'master' of github.com:scikit-learn/scikit-learn
Cosmit: explicit what OVA means as much as possible.
Cosmit
MISC: Recover changes overidden by manual merge.
BUG: Fix metrics to run on 2.5
ENH: Cosmetic improvements to the face example
MISC: Cosmit+Doc in fast truncated PCA
MISC: Remove redundant code and cosmit
Merge branch 'master' of github.com:scikit-learn/scikit-learn
ENH: Update embedded joblib to 0.4.6
ENH: import symbols on subpackage's __init__
API: Return self in _set_params
BUG: svm_gui: C is not defined for OC-SVM
ENH: Raise error when cloning bug estimators
BUG: Deal with 1D data in preprocessings.
BUG: fix cross_val and GridSearch in unsupervised.
BUG: Fix GridSearch in unsupervised
BUG: Fix the doc-generation of examples
Cosmit
MISC: Fix example
DOC: minor changes in gaussian_process docs
BUG: Fix missing gaussian_process subpackage in setup.py
FIX more missing files in setup.py
API: Remove long-depreciated function
BUG: FIx doctests broken in previous commit
DOC: documentation CD Enet fit parameters
DOC: Cosmit in docs
DOC: score is reserved to 'better is higher'
DOC: Better plotting in RFE example
ENH: Small tweak in BaseEstimator repr
ENH: Add control of the dtype in img_to_graph
ENH: dtype is img_to_graph defaults to input dtype
Mathieu Blondel (59):
Support fit_transform() in pipeline.
Support lists for training data in grid_search.
Use fit_transform and use iterables for documents.
Remove uncessary code.
Save memory when the matrix is built.
Added fit_transform() to pipeline.
Vectorizer should implement fit_transform.
SparseCountVectorizer, SparseTfidfTransformer and Sparse Vectorizer
normalize option for TfidfTransformer
fix garbage
Fix indentation.
Fix cross_val when y is a 2d-array.
Add refit option to GridSearchCV.
Merge branch 'master' into textextract
API changes to precision_recall
Fix consistency problem in the order of arguments for loss functions.
Add fbeta_score and f1_score metrics.
Rename roc to roc_curve.
Merge branch 'master' into textextract
Fix doctest in grid_search.
Merge branch 'master' into textextract
Add tests for predict_proba in LogisticRegression.
Use filter object.
Add dtype parameter to CountVectorizer and SparseCountVectorizer.
A few optimizations.
Move sparse code to sparse module.
Remove Sparse prefix from class names.
Move preprocessing to its own module.
Add Normalizer, LengthNormalizer and Binarizer.
Remove normalize option from TfidfTransformer.
Sparse equivalents of Normalizer, LengthNormalizer and Binarizer.
Fix hierarchy inconsistency for sparse module.
Move common sparse code to SparseBaseLibLinear.
Fix Sparse Logistic Regression.
Import LogisticRegression in sparse/__init__.py.
Merge branch 'master' into textextract
Activate class_weight option in fit() for liblinear-based classes.
Merge branch 'master' into textextract
Fix slicing issue when using sparse matrices.
Y -> y (capital letter is for 2d-arrays)
Raise exception when X_train.shape[1] and X_test.shape[1] don't agree.
Merge branch 'textextract' of git://github.com/ogrisel/scikit-learn into textextract
Merge branch 'textextract' of git://github.com/ogrisel/scikit-learn into textextract
Merge branch 'textextract'
Convert sparse matrix to CSR format in grid search.
Fix imports.
Pass kwargs to mlcomp loader.
Fix SGD-based binary classification example.
Note on fit_transform.
Test compute Gram matrix with support vectors only.
Activate stop word removal by default.
Add vocabulary property.
Fix small typos.
Make max_df to 1.0 by default.
Update matrix type in documentation.
Fix broken test.
class_weight="auto" for liblinear-based and sparse classes.
Fix math rendering in SVM documentation.
Fix typo.
Olivier Grisel (207):
merge textextract branch from mblondel
make analyzers inherit from BaseEstimators to get better repr and parameters management
work in progress: refactoring the document classification dataset API to remove the feature extraction step
Merge branch 'master' into textextract
tracking changes from master
cosmit
more precise doc on SVM complexity
FIX: explained variance is not symmetric: ground truth comes first
Merge branch 'textextract' of git://github.com/mblondel/scikit-learn into textextract
ENH: s/filter/preprocessor/ + docstring cosmit
ENH: more docstring love
ENH: implement __repr__ for DefaultPreprocessor so that estimators __repr__ looks prettier
cosmit
make the pipeline / grid_search object nicer to introspect in tests
FIX: make grid_search output deterministic even in case of tie on the scores
mereging pprett sgd work while tracking master
mark heisentest as skipped: it randomly passes 3 out of 5 times on my box with pyamg installed
Merge branch 'master' into sgd
kill evil tabs
cosmit
cosmit on example
cosmit: PEP8 + some missing docstrings
more cosmit
more cosmit
merging with alexandre's fixes
fix broken doctests: they are space sensitive unfortunately
Merge branch 'master' into textextract
better way to load folder / files dataset
porting the sparse document classification example to the new API
cosmit: PEP8 + some missing docstrings
Merge branch 'master' into textextract
PEP8 + better docstrings
FIX: add missing README.txt file for the sgd examples
ENH: more cosmit, docstring, test cleanup for the metrics module
cosmit
register the SGD chapter in the user guide TOC
PEP 8 in metrics module
ignore generated doc elements
some more cosmits / PEP8
more PEP8
helpers to use tags with vim / emacs
Make it possible to pass explicit labels to confusion matrix
make binary classification recall explicit
factorizing code to make it easier to do the multiclass refactoring in one place
refactored test_metrics to handle the binary case explictly and make room for
test precision recall for binary classification
more code factorization: fscore joins the party
and you thought you could escape the PEP8 screening
missing test for f1 output
cosmit
extract label extraction logics
ENH: make precision, recall and f1_score handle multi-class
removing test for multi label perf evaluation
FIX: area under curve: recall is x and precision is y
Merge branch 'master' into issue-155-multiclass-precision-recall
Merge branch 'master' into issue-155-multiclass-precision-recall
ENH: new utitity in metrics module: the classification report
showcase the new classification report in the examples
add detailed performance report to the digits example
Merge branch 'master' into issue-155-multiclass-precision-recall
tracking changes occurring on master
cosmits
ENH: handle support to do weighted averages of P/R/F scores
scalar scoring functions for P / R / Fbeta
s/explained_variance/explained_variance_score
make the distinction between loss and score function more explicit
Merge branch 'master' into textextract
spelling
make the grid search able to use an arbitrary score function
Merge branch 'master' into textextract
removing the hashing vectorizers code that need a full rewrite
update SGD example to showcase the new OVA implementation
cosmit in k_means module
FIX: better k-means tests + fixed broken array init
FIX: potential division by zero in scaler
FIX: fixed more cheesy NaNs than an Indian restaurant in Paris
New example to demonstrate the KMeans API with various init strategies
trailing spaces holocaust
let me introduce the culprit of the last checkin
Merge branch 'master' into dense
more trailing spaces cleanup
ignore downloaded data from example
Various improvement in low dim classification example
Merge branch 'dense' of git://github.com/pprett/scikit-learn into dense
remaining conflict markers in previous checkin
cosmit in SGD example
s/libsvm/liblinear/ in classification example
remove the dependency to explicit ABC to keep 2.5 compat + PEP8
make the dense SGD code & docstring more readable
more precise docstring in base SGD class
forgot to finish a sentence on regularization in a docstring
more docstring love
cosmit
use multi proc in multiclass SGD by default
cosmits in SGD tests
more comits in the sgd tests
cleanup
cosmit
better test file name
reuse the dense SGD test suite for the sparse variant using test case inheritance
cosmit
cosmits in the SGD pyx files
more cosmit in pyx files
more cosmit in example
PEP8 in SGD tests + docstring
better looking docstring for sparse sgd
more info on loss and penalty params for sparse SGD
propagate spelling fixes to the dense SGD docstring
ducktyping in analyzers
work in progress on vocabulary dimension restriction
small fixes + updated the tests
cosmit
add note on fortran contiguous memory optim for the X array
Merge branch 'master' into textextract2
OPTIM: vectorizer with predifined dictionary 5x faster by eliminating scipy.sparse.vstack calls
some optims in the text preprocessors
OPTIM: sparse vectorizer uses COO a init
multi-line print cosmit
use a SGD model in the mlcomp demo since it is the fastest for this problem
cosmit
make it possible to do fancy indexing on filenames
move the mlcomp SGD example as a generic 20 newsgroup classification example
cosmit
better pipeline notation in vectorizer + classifier grid search example
4 more years!^W^W^W 1 more test for vectorizers with max_features
cosmit
factorize out shuffling dataset since it might be useful by default
new example on how to use pipeline / grid_search for extraction parameters
sample run output in the grid_search_text_extraction_parameters example
reST formatting of example
cosmit
better title for the mlcomp example
better example filename
reference new example in the documentation of the grid_search module
cosmit
ENH: automated class_weight for SVC on imbalanced data
more s/predict_margin/decision_function/ in examples
FIX: typo in custom score_func in grid_search
initial face regonition example using eigenfaces
FIX: better handling of NaNs in precision / recall / f-score metrics
Merge branch 'faces-example'
face recognition example using eigenfaces and SVMs
more explicit subplot titles
cosmit
FIX: actually truncate the SVD to make it faster + add some test
forgot the test file in my last checkin...
drop the warning since useful even if approximate as demoed in the faces example
make fast_svd deteriministc by default while allowing to pass rng seeds
test singular values as well
new benchmark: comparing SVD implementations
remove useless import
more documentation on fast SVD + missing reference
PEP8 + various cosmits in sample generators
more tests for the iterated power refinement of the Martinsson randomized SVD
ENH: make the PCA transformer use the iterated power refinement by default
one more test for SVD
Welcome to Alexandre Passos
OPTIM: do not allocate a (n_samples, n_samples) temporary array with scipy.linalg.qr when (n_samples, k + p)) is all what is needed
OPTIM: fast_svd now has a auto transpose mode that switch to the fastest impl
cosmit
switching back to scipy.linalg.qr with econ=True to avoid half-installed numpy issues with wrong lapack bindings
FIX: numerical instability in Rdige regression tests
cosmit
new example: principal eigen / singular vector of the wikipedia graph
Better docstrings in the example
simpler SVD benchmark: use the sample_generator utility and fixed effective rank
moving real word examples to the applications subfolder
better gitignore data archives
s/_sparsedot/safe_sparse_dot/g
even better .gitignore (teasing...)
cosmit on PCA module
avoid global variable in test
ENH: make the PCA transformer perform variance scaling by default + update the face recognition accordingly
FIX: GridSearchCV refit did not propagate the fit params
switch whintening off in PCA by default + ensure unit scale + better docstring
use a grid search for the SVM params in the faces example
updated lasso benchmark to showcase the region where LassoLARS is faster than Lasso CD
OPTIM: ensure lasso_path aligns the data only once in if not alread fortran contiguous
pep8
ENH: LassoCV / ElasticNetCV now uses all folds data + example
ENH: make the LassoLARS and LassoCD path examples easier to compare
make MSE plot of LassoCV more readable by scaling the y axis
FIX: update broken tests by last checkin
switch to base 10 for the alpha logs in the Lasso CD path plot
revert the plot style to the LARS paper conventions
select the best alpha using the mean of the CV MSEs instead of the median
cosmit: += assignement replaced by plain = in coorinate_decent (more natural, less confusing)
extract the randomized SVD implementation as a toplevel class able to handle sparse data as well
consistently rename n_comp to n_components
Merge branch 'master' into sparse-pca
update doctest to handle the change in regularizer strenght definition in LARS
FIX: typo s/mean/mean_/g in RandomizedPCA
Merge branch 'master' into sparse-pca
sed -i "s/\<n_componentsonents\>/n_components/g"
SVD benchmark have a consistent filename
factorized out correlated regression dataset utility function and updated
do not allocate useless memory in make_regression_dataset
launch test on documentation by default when running make
cosmit
OPTIM: do not precompute r2_score_ in ElasticNet in the fit call
do not precompute explained_variance_ in linear model: can be too costly: use r2_score when needed instead
new benchmark for lasso path implementations
merging master
temporary test fix for refit instability in linear SVC: a bugfix branch will be open to reproduce the issue
cosmit (reST formatting of the SGD module documentation)
Merge branch 'master' of github.com:scikit-learn/scikit-learn
more formatting in SGD reST and fixed docstest broken by last checkin :(
Peter Prettenhofer (97):
initial checkin of sgd package.
set rho on 1 or 0 if L2 or L1 penalty.
l1 penalty implemented.
added class encoding.
does not belong to the repo.
Merge branch 'master' of git at github.com:pprett/scikit-learn
Code review from Alexandre:
Merge branch 'master' of github.com:pprett/scikit-learn
removed unnecessary print statements.
100% code coverage.
added doctests to SGD and sgd.LinearModel
initial checkin of sgd package.
set rho on 1 or 0 if L2 or L1 penalty.
l1 penalty implemented.
added class encoding.
Code review from Alexandre:
100% code coverage.
added doctests to SGD and sgd.LinearModel
Merge commit 'origin/master'
initial *draft* of the sgd module documentation added.
added Readme so that sphinx stops complaining.
additional documentation for sgd (plot of various convex loss functions).
math formulation cont'
penalty contour plot added.
more SGD documentation added: example, math formulation , implementation details.
EfficientBackprop reference added.
Documentation for sgd polished.
fixed doctests after SGD class index refactoring.
Removed tabs.
implemented OVA for multi-class SGD.
implemented OVA for multi-class SGD.
Merge branch 'master' into ova
SGD supports multi-class classification using one-vs.-all.
SGD multi-class documentation added.
SGD classifier supports multi-class with OVA.
documentation for multi-class sgd updated.
Changed docstrings for coef_ and intercept_ in sgd package. Wrap intercept_ in an array in the case of binary classification.
Merge branch 'master' of git at github.com:scikit-learn/scikit-learn
Liblinear docstring modified: deleted irrelevant attributes support_ and changed shape of intercept_ and coef_ accordingly.
Added dense implemenation of SGD.
Merge branch 'dense' of github.com:pprett/scikit-learn
Commit broken cython header import.
moved sgd_fast_sparse from sgd/sparse/src to sgd/src.
Moved sgd extension modules from sgd/src to sgd.
performance improvements in cython files; cython files rebuild.
added covertype example for dense sgd.
bugfix in plot_loss_functions (import loss functions).
covertype example now downloads dataset automatically.
Updated sgd documentation with multi-class documentation.
docstrings: n_jobs defaults to 1.
cosmit: color of data points matches color of decision regions and OVA hyperplanes.
warm start optimization changed from coef_ to init_coef_ and intercept_ to init_intercept_.
Multi-class documentation for module sgd added.
Include models with L1 and Elastic-Net penalty.
changed init_coef to coef_init (intercept likewise).
Merge branch 'warmstart'
Added new example on modeling the geographic distribution of species.
Merge branch 'speciesmodeling' of git at github.com:pprett/scikit-learn
added species distribution example as plot example.
if possible, species distribution example now uses basemap by default.
deleted old species_distribution_modeling example.
cosmit: pep8 and author
Reduced memory consumption in covertype example due to memory leak in np.loadtxt.
Merge branch 'master' of github.com:scikit-learn/scikit-learn
Added note on the importance of shuffeling. Minor changes in text.
Runtime improvement of species distribution example (fancy indexing).
set basemap as default.
Class weights for SGD similar to svm package. Same heuristic as Liblinear for multi-class (OVA): use only weight for the positive class.
removed parameters `p` and `C` from OneClassSVM (dense and sparse).
Merge branch 'master' of git at github.com:scikit-learn/scikit-learn
added tksvm from git://gist.github.com/673953.git.
use np.fromstring to load data from large csv text files.
changed predict_margin to decision_function
Merge branch 'svmgui'
added GUI example for SVM.
added tksvm from git://gist.github.com/673953.git.
added GUI example for SVM.
Merge branch 'master' of git at github.com:scikit-learn/scikit-learn
Merge branch 'importanceweighting' of git at github.com:pprett/scikit-learn
RegressorSGD added.
Merge branch 'master' into importanceweighting
changend "squarederror" to "squaredloss".
automatic refitting on radiobutton change and add example.
Merge branch 'master' of git at github.com:scikit-learn/scikit-learn
changed loss function names in SGD (squaredloss -> squared_loss; also for modified_huber).
added Oliviers ElasticNet convergence test to SGD.
move sgd into linear_model and rename sgd to stochastic_gradient.
finalized sgd module renaming.
moved sgd examples to examples/linear_model and added sgd prefix.
Merge branch 'master' of git at github.com:scikit-learn/scikit-learn
Merge branch 'master' into sgd-rename
COSMIT: smaller data points
updated SGD documentation (referenced in linear_model.rst and classes.rst).
fixed imports in non-auto examples.
BUGFIX in sparse.SGDRegressor
Merge branch 'master' of git at github.com:scikit-learn/scikit-learn
Merge branch 'sgd-rename'
Ron Weiss (21):
change GMM initialization to use cluster.KMeans
change GaussianHMM initialization to use cluster.KMeans
merge
fix bug in hmm.GaussianHMM mstep update for 'full' covariance
Reapply "ENH: enhacements in the gmm module."
fix gmm examples
merge
fix bug in GMM._get_covars dimensions
make HMM interface consistent with GMM
clean up interfaces in hmm and gmm
remove n_symbols argument from MultinomialHMM.__init__
Merge branch 'master' of github.com:scikit-learn/scikit-learn
add GMM classification example
clarify GMM classifier labels
add GMM.predict_proba
add default initialization of GMM.weights to constructor
rename GMM.n_dim to GMM.n_features to be consistent with the rest of the scikit
add HMM.predict_proba
rename HMM.n_dim to HMM.n_features to be consistent with the rest of the scikit
fix pep8, pyflakes errprs
Merge branch 'master' of github.com:scikit-learn/scikit-learn
Ronan Amicel (2):
Fix broken merge by ogrisel :-P
PEP8 + missing fit methods
Vincent Dubourg (33):
Hello list,
Correction of a bug with the management of the dimension of the autocorrelation parameters.
Forgot to retire pdb.
Commit of a 'Gaussian Process for Machine Learning' module in the gpml directory. The module implement a class named GaussianProcessModel. I also add doc, examples and tests (involving a coupling with the cross_val module).
Correction of a bug in test_gpml.py (now runs perfect on my machine!). I just don't know how to involve this test within the whole scikit testing procedure (nosetests). Also add a modification of the TOC in doc.
Correction of a bug in the basic regression example.
Delete the old kriging.py module
Modification of the score function. The score function now evaluates the deviation between the predicted targets and the true ones. This is for convenience only because it allows then to use the distributing capacity of the cross_val module. The old score function is renamed with the more explicit name: `reduced_likelihood_function` (see eg the DACE documentation).
Modification of the main __init__.py file of the scikits.learn package in order to load the gpml module and tests.
Renames as suggested by Alexandre. Simplification of the examples. Remove the interactive contour label picking in the probabilistic classification example.
Bugged example after modification. Now correct!
I Ran the PEP8 and PYFLAKES utils and corrected the gaussian_process module related files.
Can't comply with contradictory PEP8 rules on some specfic code such as:
I removed the time-consuming test and made a regression example from it.
Replaced np.matrix(A) * np.matrix(B) by np.dot(A,B), so that the code is a lot clearer to read...
Removed plotting command from the examples in the GaussianProcess class docstring.
Simplification of input's shape checking using np.atleast_2d()
Changes in format of the fit() input (np.atleast_2d for X, and np.newaxis cat for y).
Force y to np.array before concatenating np.newaxis in fit().
Modifications following Gaël latest remarks.
Added Welch's MLE optimizer in arg_max_reduced_likelihood_function() plus reference in the docstring.
Correction of a minor typo error in correlation_models docstring
Improvement of the documentation with a piece of code and reference to the regression auto_example. Add a README.txt file at the root of the examples/gaussian_process directory.
From: agramfort: don't use capital letters for a vector. Y -> y.
Forgot to retire pdb... Again!
Forgot one capital Y in the piece of code of the RST docpage.
Removed trailing spaces in the RST doc page.
Merge branch 'master' of https://github.com/scikit-learn/scikit-learn
metrics.explained_variance was renamed to metrics.explained_variance_score so that I needed to modify this example.
Removal of the submodule relative imports in the toplevel init file.
gaussian_process module changes:
Merge branch 'master' of github.com:scikit-learn/scikit-learn
Merge branch 'gaussian_process_review'
Yaroslav Halchenko (11):
Merge 0.6.0 into DFSG (reincarnating libsvm's svm.* -- effectively forked)
Merge branch 'dfsg' into debian
Disabled use of system libsvm - scikits.learn has its own fork
Updated copyright: DEP5 rev153, entries for cblas/ and lib{linear/svm}
Extended TODO (in emacs-org mode): consider system's liblinear (need 1.7 with a fix)
changelog entry for 0.6.dfsg-1
Added README.source to describe reason(s) for .dfsg
debian/watch: mangle debian version - remove .dfsg for comparisons
Fixed upstream release number -- 0.6.0, not 0.6
adjusting dfsg rule -- do not prune libsvm
Extending changelog
-----------------------------------------------------------------------
No new revisions were added by this update.
--
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/debian-science/packages/scikit-learn.git
More information about the debian-science-commits
mailing list