scikit-learn/sklearn/svm/sparse/base.py

import numpy as np
import scipy.sparse
from abc import ABCMeta, abstractmethod

from ..base import BaseLibSVM


class SparseBaseLibSVM(BaseLibSVM):
    __metaclass__ = ABCMeta

    @abstractmethod
    def __init__(self, impl, kernel, degree, gamma, coef0,
                 tol, C, nu, epsilon, shrinking, probability, cache_size,
                 class_weight, verbose, max_iter):

        super(SparseBaseLibSVM, self).__init__(impl, kernel, degree, gamma,
                coef0, tol, C, nu, epsilon, shrinking, probability, cache_size,
                True, class_weight, verbose, max_iter)

    def fit(self, X, y, sample_weight=None):
        X = scipy.sparse.csr_matrix(X, dtype=np.float64)
        return super(SparseBaseLibSVM, self).fit(X, y,
                                                 sample_weight=sample_weight)
move sparse.svm into the svm module to match glm.sparse. 2010-09-28 20:04:29 +08:00			`import numpy as np`
FIX: do not use reshape in libsvm sparse bindings. This should fix issues of people using this in parallel. 2011-11-07 18:03:14 +08:00			`import scipy.sparse`
Make sparse svm base class ABC 2012-06-08 03:38:11 +08:00			`from abc import ABCMeta, abstractmethod`
move sparse.svm into the svm module to match glm.sparse. 2010-09-28 20:04:29 +08:00
ENH merge sparse and dense SVMs, part 1 * enable sparse matrix support in formerly dense versions * remove DenseBaseLibSVM * move sparse.libsvm to libsvm_sparse 2012-01-20 06:39:06 +08:00			`from ..base import BaseLibSVM`
Move common sparse code to SparseBaseLibLinear. 2010-10-26 13:51:54 +08:00
pep8 stuff as well removed testing stuff 2011-02-06 02:27:03 +08:00
move sparse.svm into the svm module to match glm.sparse. 2010-09-28 20:04:29 +08:00			`class SparseBaseLibSVM(BaseLibSVM):`
Make sparse svm base class ABC 2012-06-08 03:38:11 +08:00			`__metaclass__ = ABCMeta`

			`@abstractmethod`
Move cache_size out of model parameters. 2011-05-13 15:44:03 +08:00			`def __init__(self, impl, kernel, degree, gamma, coef0,`
ENH : adding support for scaling by n_samples of the C parameter in libsvm and liblinear models 2011-11-12 05:43:03 +08:00			`tol, C, nu, epsilon, shrinking, probability, cache_size,`
ENH: renaming iter_limit -> max_iter 2012-10-06 06:41:43 +08:00			`class_weight, verbose, max_iter):`
FIX: sparse svms do not accept callable kernels. 2010-12-07 19:18:07 +08:00
FIX Issue 379 and use the opportunity to refactor libsvm code Sparse SVC's would try to use the dense decision_function, since SparseBaseLibSVM did not override that method. Solution: factor out non-common code to DenseBaseLibSVM. (@diogojc's test program still crashes, but with an appropriate error message.) Also, leverage superclass __init__ in SparseBaseLibSVM 2011-10-09 10:11:10 +08:00			`super(SparseBaseLibSVM, self).__init__(impl, kernel, degree, gamma,`
Merge remote-tracking branch 'origin/master' into n_samples_scaling Conflicts: sklearn/svm/sparse/base.py sklearn/svm/tests/test_sparse.py 2011-12-18 19:45:58 +08:00			`coef0, tol, C, nu, epsilon, shrinking, probability, cache_size,`
ENH: renaming iter_limit -> max_iter 2012-10-06 06:41:43 +08:00			`True, class_weight, verbose, max_iter)`
move sparse.svm into the svm module to match glm.sparse. 2010-09-28 20:04:29 +08:00
ENH: moved class_weight parameter in svms from fit to ``__init__``. 2012-02-07 07:24:02 +08:00			`def fit(self, X, y, sample_weight=None):`
ENH merge sparse and dense SVMs, part 1 * enable sparse matrix support in formerly dense versions * remove DenseBaseLibSVM * move sparse.libsvm to libsvm_sparse 2012-01-20 06:39:06 +08:00			`X = scipy.sparse.csr_matrix(X, dtype=np.float64)`
FIX sparse OneClassSVM was using the wrong parameter 2012-09-03 02:42:50 +08:00			`return super(SparseBaseLibSVM, self).fit(X, y,`
			`sample_weight=sample_weight)`