480 lines
18 KiB
Python
480 lines
18 KiB
Python
from ctypes import *
|
|
from ctypes.util import find_library
|
|
from os import path
|
|
from glob import glob
|
|
import sys
|
|
from enum import IntEnum
|
|
try:
|
|
import numpy as np
|
|
import scipy
|
|
from scipy import sparse
|
|
except:
|
|
scipy = None
|
|
|
|
if sys.version_info[0] < 3:
|
|
range = xrange
|
|
from itertools import izip as zip
|
|
|
|
__all__ = ['liblinear', 'feature_node', 'gen_feature_nodearray', 'problem',
|
|
'parameter', 'model', 'toPyModel', 'solver_names',
|
|
'print_null']
|
|
|
|
try:
|
|
dirname = path.dirname(path.abspath(__file__))
|
|
dynamic_lib_name = 'clib.cp*'
|
|
path_to_so = glob(path.join(dirname, dynamic_lib_name))[0]
|
|
liblinear = CDLL(path_to_so)
|
|
except:
|
|
try :
|
|
if sys.platform == 'win32':
|
|
liblinear = CDLL(path.join(dirname, r'..\..\windows\liblinear.dll'))
|
|
else:
|
|
liblinear = CDLL(path.join(dirname, '../../liblinear.so.6'))
|
|
except:
|
|
# For unix the prefix 'lib' is not considered.
|
|
if find_library('linear'):
|
|
liblinear = CDLL(find_library('linear'))
|
|
elif find_library('liblinear'):
|
|
liblinear = CDLL(find_library('liblinear'))
|
|
else:
|
|
raise Exception('LIBLINEAR library not found.')
|
|
|
|
class solver_names(IntEnum):
|
|
L2R_LR = 0
|
|
L2R_L2LOSS_SVC_DUAL = 1
|
|
L2R_L2LOSS_SVC = 2
|
|
L2R_L1LOSS_SVC_DUAL = 3
|
|
MCSVM_CS = 4
|
|
L1R_L2LOSS_SVC = 5
|
|
L1R_LR = 6
|
|
L2R_LR_DUAL = 7
|
|
L2R_L2LOSS_SVR = 11
|
|
L2R_L2LOSS_SVR_DUAL = 12
|
|
L2R_L1LOSS_SVR_DUAL = 13
|
|
ONECLASS_SVM = 21
|
|
|
|
PRINT_STRING_FUN = CFUNCTYPE(None, c_char_p)
|
|
def print_null(s):
|
|
return
|
|
|
|
# In multi-threading, all threads share the same memory space of
|
|
# the dynamic library (liblinear). Thus, we use a module-level
|
|
# variable to keep a reference to ctypes print_null, preventing
|
|
# python from garbage collecting it in thread B while thread A
|
|
# still needs it. Check the usage of svm_set_print_string_function()
|
|
# in LIBLINEAR README for details.
|
|
ctypes_print_null = PRINT_STRING_FUN(print_null)
|
|
|
|
def genFields(names, types):
|
|
return list(zip(names, types))
|
|
|
|
def fillprototype(f, restype, argtypes):
|
|
f.restype = restype
|
|
f.argtypes = argtypes
|
|
|
|
class feature_node(Structure):
|
|
_names = ["index", "value"]
|
|
_types = [c_int, c_double]
|
|
_fields_ = genFields(_names, _types)
|
|
|
|
def __str__(self):
|
|
return '%d:%g' % (self.index, self.value)
|
|
|
|
def gen_feature_nodearray(xi, feature_max=None):
|
|
if feature_max:
|
|
assert(isinstance(feature_max, int))
|
|
|
|
xi_shift = 0 # ensure correct indices of xi
|
|
if scipy and isinstance(xi, tuple) and len(xi) == 2\
|
|
and isinstance(xi[0], np.ndarray) and isinstance(xi[1], np.ndarray): # for a sparse vector
|
|
index_range = xi[0] + 1 # index starts from 1
|
|
if feature_max:
|
|
index_range = index_range[np.where(index_range <= feature_max)]
|
|
elif scipy and isinstance(xi, np.ndarray):
|
|
xi_shift = 1
|
|
index_range = xi.nonzero()[0] + 1 # index starts from 1
|
|
if feature_max:
|
|
index_range = index_range[np.where(index_range <= feature_max)]
|
|
elif isinstance(xi, (dict, list, tuple)):
|
|
if isinstance(xi, dict):
|
|
index_range = sorted(xi.keys())
|
|
elif isinstance(xi, (list, tuple)):
|
|
xi_shift = 1
|
|
index_range = range(1, len(xi) + 1)
|
|
index_range = list(filter(lambda j: xi[j-xi_shift] != 0, index_range))
|
|
|
|
if feature_max:
|
|
index_range = list(filter(lambda j: j <= feature_max, index_range))
|
|
else:
|
|
raise TypeError('xi should be a dictionary, list, tuple, 1-d numpy array, or tuple of (index, data)')
|
|
|
|
ret = (feature_node*(len(index_range)+2))()
|
|
ret[-1].index = -1 # for bias term
|
|
ret[-2].index = -1
|
|
|
|
if scipy and isinstance(xi, tuple) and len(xi) == 2\
|
|
and isinstance(xi[0], np.ndarray) and isinstance(xi[1], np.ndarray): # for a sparse vector
|
|
# since xi=(indices, values), we must sort them simultaneously.
|
|
for idx, arg in enumerate(np.argsort(index_range)):
|
|
ret[idx].index = index_range[arg]
|
|
ret[idx].value = (xi[1])[arg]
|
|
else:
|
|
for idx, j in enumerate(index_range):
|
|
ret[idx].index = j
|
|
ret[idx].value = xi[j - xi_shift]
|
|
|
|
max_idx = 0
|
|
if len(index_range) > 0:
|
|
max_idx = index_range[-1]
|
|
return ret, max_idx
|
|
|
|
try:
|
|
from numba import jit
|
|
jit_enabled = True
|
|
except:
|
|
# We need to support two cases: when jit is called with no arguments, and when jit is called with
|
|
# a keyword argument.
|
|
def jit(func=None, *args, **kwargs):
|
|
if func is None:
|
|
# This handles the case where jit is used with parentheses: @jit(nopython=True)
|
|
return lambda x: x
|
|
else:
|
|
# This handles the case where jit is used without parentheses: @jit
|
|
return func
|
|
jit_enabled = False
|
|
|
|
@jit(nopython=True)
|
|
def csr_to_problem_jit(l, x_val, x_ind, x_rowptr, prob_val, prob_ind, prob_rowptr):
|
|
for i in range(l):
|
|
b1,e1 = x_rowptr[i], x_rowptr[i+1]
|
|
b2,e2 = prob_rowptr[i], prob_rowptr[i+1]-2
|
|
for j in range(b1,e1):
|
|
prob_ind[j-b1+b2] = x_ind[j]+1
|
|
prob_val[j-b1+b2] = x_val[j]
|
|
def csr_to_problem_nojit(l, x_val, x_ind, x_rowptr, prob_val, prob_ind, prob_rowptr):
|
|
for i in range(l):
|
|
x_slice = slice(x_rowptr[i], x_rowptr[i+1])
|
|
prob_slice = slice(prob_rowptr[i], prob_rowptr[i+1]-2)
|
|
prob_ind[prob_slice] = x_ind[x_slice]+1
|
|
prob_val[prob_slice] = x_val[x_slice]
|
|
|
|
def csr_to_problem(x, prob):
|
|
if not x.has_sorted_indices:
|
|
x.sort_indices()
|
|
|
|
# Extra space for termination node and (possibly) bias term
|
|
x_space = prob.x_space = np.empty((x.nnz+x.shape[0]*2), dtype=feature_node)
|
|
# rowptr has to be a 64bit integer because it will later be used for pointer arithmetic,
|
|
# which overflows when the added pointer points to an address that is numerically high.
|
|
prob.rowptr = x.indptr.astype(np.int64, copy=True)
|
|
prob.rowptr[1:] += 2*np.arange(1,x.shape[0]+1)
|
|
prob_ind = x_space["index"]
|
|
prob_val = x_space["value"]
|
|
prob_ind[:] = -1
|
|
if jit_enabled:
|
|
csr_to_problem_jit(x.shape[0], x.data, x.indices, x.indptr, prob_val, prob_ind, prob.rowptr)
|
|
else:
|
|
csr_to_problem_nojit(x.shape[0], x.data, x.indices, x.indptr, prob_val, prob_ind, prob.rowptr)
|
|
|
|
class problem(Structure):
|
|
_names = ["l", "n", "y", "x", "bias"]
|
|
_types = [c_int, c_int, POINTER(c_double), POINTER(POINTER(feature_node)), c_double]
|
|
_fields_ = genFields(_names, _types)
|
|
|
|
def __init__(self, y, x, bias = -1):
|
|
if (not isinstance(y, (list, tuple))) and (not (scipy and isinstance(y, np.ndarray))):
|
|
raise TypeError("type of y: {0} is not supported!".format(type(y)))
|
|
|
|
if isinstance(x, (list, tuple)):
|
|
if len(y) != len(x):
|
|
raise ValueError("len(y) != len(x)")
|
|
elif scipy != None and isinstance(x, (np.ndarray, sparse.spmatrix)):
|
|
if len(y) != x.shape[0]:
|
|
raise ValueError("len(y) != len(x)")
|
|
if isinstance(x, np.ndarray):
|
|
x = np.ascontiguousarray(x) # enforce row-major
|
|
if isinstance(x, sparse.spmatrix):
|
|
x = x.tocsr()
|
|
pass
|
|
else:
|
|
raise TypeError("type of x: {0} is not supported!".format(type(x)))
|
|
self.l = l = len(y)
|
|
self.bias = -1
|
|
|
|
max_idx = 0
|
|
x_space = self.x_space = []
|
|
if scipy != None and isinstance(x, sparse.csr_matrix):
|
|
csr_to_problem(x, self)
|
|
max_idx = x.shape[1]
|
|
else:
|
|
for i, xi in enumerate(x):
|
|
tmp_xi, tmp_idx = gen_feature_nodearray(xi)
|
|
x_space += [tmp_xi]
|
|
max_idx = max(max_idx, tmp_idx)
|
|
self.n = max_idx
|
|
|
|
self.y = (c_double * l)()
|
|
if scipy != None and isinstance(y, np.ndarray):
|
|
np.ctypeslib.as_array(self.y, (self.l,))[:] = y
|
|
else:
|
|
for i, yi in enumerate(y): self.y[i] = yi
|
|
|
|
self.x = (POINTER(feature_node) * l)()
|
|
if scipy != None and isinstance(x, sparse.csr_matrix):
|
|
base = addressof(self.x_space.ctypes.data_as(POINTER(feature_node))[0])
|
|
x_ptr = cast(self.x, POINTER(c_uint64))
|
|
x_ptr = np.ctypeslib.as_array(x_ptr,(self.l,))
|
|
x_ptr[:] = self.rowptr[:-1]*sizeof(feature_node)+base
|
|
else:
|
|
for i, xi in enumerate(self.x_space): self.x[i] = xi
|
|
|
|
self.set_bias(bias)
|
|
|
|
def set_bias(self, bias):
|
|
if self.bias == bias:
|
|
return
|
|
if bias >= 0 and self.bias < 0:
|
|
self.n += 1
|
|
node = feature_node(self.n, bias)
|
|
if bias < 0 and self.bias >= 0:
|
|
self.n -= 1
|
|
node = feature_node(-1, bias)
|
|
|
|
if isinstance(self.x_space, list):
|
|
for xi in self.x_space:
|
|
xi[-2] = node
|
|
else:
|
|
self.x_space["index"][self.rowptr[1:]-2] = node.index
|
|
self.x_space["value"][self.rowptr[1:]-2] = node.value
|
|
|
|
self.bias = bias
|
|
|
|
def copy(self):
|
|
prob_copy = problem.__new__(problem)
|
|
for key in problem._names + list(vars(self)):
|
|
setattr(prob_copy, key, getattr(self, key))
|
|
return prob_copy
|
|
|
|
|
|
class parameter(Structure):
|
|
_names = ["solver_type", "eps", "C", "nr_weight", "weight_label",
|
|
"weight", "p", "nu", "init_sol", "regularize_bias",
|
|
"w_recalc"]
|
|
_types = [c_int, c_double, c_double, c_int, POINTER(c_int),
|
|
POINTER(c_double), c_double, c_double, POINTER(c_double), c_int,c_bool]
|
|
_fields_ = genFields(_names, _types)
|
|
|
|
def __init__(self, options = None):
|
|
if options == None:
|
|
options = ''
|
|
self.parse_options(options)
|
|
|
|
def __str__(self):
|
|
s = ''
|
|
attrs = parameter._names + list(self.__dict__.keys())
|
|
values = map(lambda attr: getattr(self, attr), attrs)
|
|
for attr, val in zip(attrs, values):
|
|
s += (' %s: %s\n' % (attr, val))
|
|
s = s.strip()
|
|
|
|
return s
|
|
|
|
def set_to_default_values(self):
|
|
self.solver_type = solver_names.L2R_L2LOSS_SVC_DUAL
|
|
self.eps = float('inf')
|
|
self.C = 1
|
|
self.p = 0.1
|
|
self.nu = 0.5
|
|
self.nr_weight = 0
|
|
self.weight_label = None
|
|
self.weight = None
|
|
self.init_sol = None
|
|
self.bias = -1
|
|
self.regularize_bias = 1
|
|
self.w_recalc = False
|
|
self.flag_cross_validation = False
|
|
self.flag_C_specified = False
|
|
self.flag_p_specified = False
|
|
self.flag_solver_specified = False
|
|
self.flag_find_parameters = False
|
|
self.nr_fold = 0
|
|
self.print_func = cast(None, PRINT_STRING_FUN)
|
|
|
|
def parse_options(self, options):
|
|
if isinstance(options, list):
|
|
argv = options
|
|
elif isinstance(options, str):
|
|
argv = options.split()
|
|
else:
|
|
raise TypeError("arg 1 should be a list or a str.")
|
|
self.set_to_default_values()
|
|
self.print_func = cast(None, PRINT_STRING_FUN)
|
|
weight_label = []
|
|
weight = []
|
|
|
|
i = 0
|
|
while i < len(argv) :
|
|
if argv[i] == "-s":
|
|
i = i + 1
|
|
self.solver_type = solver_names(int(argv[i]))
|
|
self.flag_solver_specified = True
|
|
elif argv[i] == "-c":
|
|
i = i + 1
|
|
self.C = float(argv[i])
|
|
self.flag_C_specified = True
|
|
elif argv[i] == "-p":
|
|
i = i + 1
|
|
self.p = float(argv[i])
|
|
self.flag_p_specified = True
|
|
elif argv[i] == "-n":
|
|
i = i + 1
|
|
self.nu = float(argv[i])
|
|
elif argv[i] == "-e":
|
|
i = i + 1
|
|
self.eps = float(argv[i])
|
|
elif argv[i] == "-B":
|
|
i = i + 1
|
|
self.bias = float(argv[i])
|
|
elif argv[i] == "-v":
|
|
i = i + 1
|
|
self.flag_cross_validation = 1
|
|
self.nr_fold = int(argv[i])
|
|
if self.nr_fold < 2 :
|
|
raise ValueError("n-fold cross validation: n must >= 2")
|
|
elif argv[i].startswith("-w"):
|
|
i = i + 1
|
|
self.nr_weight += 1
|
|
weight_label += [int(argv[i-1][2:])]
|
|
weight += [float(argv[i])]
|
|
elif argv[i] == "-q":
|
|
self.print_func = ctypes_print_null
|
|
elif argv[i] == "-C":
|
|
self.flag_find_parameters = True
|
|
elif argv[i] == "-R":
|
|
self.regularize_bias = 0
|
|
else:
|
|
raise ValueError("Wrong options")
|
|
i += 1
|
|
|
|
liblinear.set_print_string_function(self.print_func)
|
|
self.weight_label = (c_int*self.nr_weight)()
|
|
self.weight = (c_double*self.nr_weight)()
|
|
for i in range(self.nr_weight):
|
|
self.weight[i] = weight[i]
|
|
self.weight_label[i] = weight_label[i]
|
|
|
|
# default solver for parameter selection is L2R_L2LOSS_SVC
|
|
if self.flag_find_parameters:
|
|
if not self.flag_cross_validation:
|
|
self.nr_fold = 5
|
|
if not self.flag_solver_specified:
|
|
self.solver_type = solver_names.L2R_L2LOSS_SVC
|
|
self.flag_solver_specified = True
|
|
elif self.solver_type not in [solver_names.L2R_LR, solver_names.L2R_L2LOSS_SVC, solver_names.L2R_L2LOSS_SVR]:
|
|
raise ValueError("Warm-start parameter search only available for -s 0, -s 2 and -s 11")
|
|
|
|
if self.eps == float('inf'):
|
|
if self.solver_type in [solver_names.L2R_LR, solver_names.L2R_L2LOSS_SVC]:
|
|
self.eps = 0.01
|
|
elif self.solver_type in [solver_names.L2R_L2LOSS_SVR]:
|
|
self.eps = 0.0001
|
|
elif self.solver_type in [solver_names.L2R_L2LOSS_SVC_DUAL, solver_names.L2R_L1LOSS_SVC_DUAL, solver_names.MCSVM_CS, solver_names.L2R_LR_DUAL]:
|
|
self.eps = 0.1
|
|
elif self.solver_type in [solver_names.L1R_L2LOSS_SVC, solver_names.L1R_LR]:
|
|
self.eps = 0.01
|
|
elif self.solver_type in [solver_names.L2R_L2LOSS_SVR_DUAL, solver_names.L2R_L1LOSS_SVR_DUAL]:
|
|
self.eps = 0.1
|
|
elif self.solver_type in [solver_names.ONECLASS_SVM]:
|
|
self.eps = 0.01
|
|
|
|
class model(Structure):
|
|
_names = ["param", "nr_class", "nr_feature", "w", "label", "bias", "rho"]
|
|
_types = [parameter, c_int, c_int, POINTER(c_double), POINTER(c_int), c_double, c_double]
|
|
_fields_ = genFields(_names, _types)
|
|
|
|
def __init__(self):
|
|
self.__createfrom__ = 'python'
|
|
|
|
def __del__(self):
|
|
# free memory created by C to avoid memory leak
|
|
if hasattr(self, '__createfrom__') and self.__createfrom__ == 'C':
|
|
liblinear.free_and_destroy_model(pointer(self))
|
|
|
|
def get_nr_feature(self):
|
|
return liblinear.get_nr_feature(self)
|
|
|
|
def get_nr_class(self):
|
|
return liblinear.get_nr_class(self)
|
|
|
|
def get_labels(self):
|
|
nr_class = self.get_nr_class()
|
|
labels = (c_int * nr_class)()
|
|
liblinear.get_labels(self, labels)
|
|
return labels[:nr_class]
|
|
|
|
def get_decfun_coef(self, feat_idx, label_idx=0):
|
|
return liblinear.get_decfun_coef(self, feat_idx, label_idx)
|
|
|
|
def get_decfun_bias(self, label_idx=0):
|
|
return liblinear.get_decfun_bias(self, label_idx)
|
|
|
|
def get_decfun_rho(self):
|
|
return liblinear.get_decfun_rho(self)
|
|
|
|
def get_decfun(self, label_idx=0):
|
|
w = [liblinear.get_decfun_coef(self, feat_idx, label_idx) for feat_idx in range(1, self.nr_feature+1)]
|
|
if self.is_oneclass_model():
|
|
rho = self.get_decfun_rho()
|
|
return (w, -rho)
|
|
else:
|
|
b = liblinear.get_decfun_bias(self, label_idx)
|
|
return (w, b)
|
|
|
|
def is_probability_model(self):
|
|
return (liblinear.check_probability_model(self) == 1)
|
|
|
|
def is_regression_model(self):
|
|
return (liblinear.check_regression_model(self) == 1)
|
|
|
|
def is_oneclass_model(self):
|
|
return (liblinear.check_oneclass_model(self) == 1)
|
|
|
|
def toPyModel(model_ptr):
|
|
"""
|
|
toPyModel(model_ptr) -> model
|
|
|
|
Convert a ctypes POINTER(model) to a Python model
|
|
"""
|
|
if bool(model_ptr) == False:
|
|
raise ValueError("Null pointer")
|
|
m = model_ptr.contents
|
|
m.__createfrom__ = 'C'
|
|
return m
|
|
|
|
fillprototype(liblinear.train, POINTER(model), [POINTER(problem), POINTER(parameter)])
|
|
fillprototype(liblinear.find_parameters, None, [POINTER(problem), POINTER(parameter), c_int, c_double, c_double, POINTER(c_double), POINTER(c_double), POINTER(c_double)])
|
|
fillprototype(liblinear.cross_validation, None, [POINTER(problem), POINTER(parameter), c_int, POINTER(c_double)])
|
|
|
|
fillprototype(liblinear.predict_values, c_double, [POINTER(model), POINTER(feature_node), POINTER(c_double)])
|
|
fillprototype(liblinear.predict, c_double, [POINTER(model), POINTER(feature_node)])
|
|
fillprototype(liblinear.predict_probability, c_double, [POINTER(model), POINTER(feature_node), POINTER(c_double)])
|
|
|
|
fillprototype(liblinear.save_model, c_int, [c_char_p, POINTER(model)])
|
|
fillprototype(liblinear.load_model, POINTER(model), [c_char_p])
|
|
|
|
fillprototype(liblinear.get_nr_feature, c_int, [POINTER(model)])
|
|
fillprototype(liblinear.get_nr_class, c_int, [POINTER(model)])
|
|
fillprototype(liblinear.get_labels, None, [POINTER(model), POINTER(c_int)])
|
|
fillprototype(liblinear.get_decfun_coef, c_double, [POINTER(model), c_int, c_int])
|
|
fillprototype(liblinear.get_decfun_bias, c_double, [POINTER(model), c_int])
|
|
fillprototype(liblinear.get_decfun_rho, c_double, [POINTER(model)])
|
|
|
|
fillprototype(liblinear.free_model_content, None, [POINTER(model)])
|
|
fillprototype(liblinear.free_and_destroy_model, None, [POINTER(POINTER(model))])
|
|
fillprototype(liblinear.destroy_param, None, [POINTER(parameter)])
|
|
fillprototype(liblinear.check_parameter, c_char_p, [POINTER(problem), POINTER(parameter)])
|
|
fillprototype(liblinear.check_probability_model, c_int, [POINTER(model)])
|
|
fillprototype(liblinear.check_regression_model, c_int, [POINTER(model)])
|
|
fillprototype(liblinear.check_oneclass_model, c_int, [POINTER(model)])
|
|
fillprototype(liblinear.set_print_string_function, None, [CFUNCTYPE(None, c_char_p)])
|