def __init__(self, y, x):
assert len(y) == len(x)
self.prob = prob = svmc.new_svm_problem()
self.size = size = len(y)
self.y_array = y_array = svmc.new_double(size)
for i in range(size):
svmc.double_setitem(y_array, i, y[i])
self.x_matrix = x_matrix = svmc.svm_node_matrix(size)
self.data = []
self.maxlen = 0
for i in range(size):
data = _convert_to_svm_node_array(x[i])
self.data.append(data)
svmc.svm_node_matrix_set(x_matrix, i, data)
if type(x[i]) == dict:
if (len(x[i]) > 0):
self.maxlen = max(self.maxlen, max(x[i].keys()))
else:
self.maxlen = max(self.maxlen, len(x[i]))
svmc.svm_problem_l_set(prob, size)
svmc.svm_problem_y_set(prob, y_array)
svmc.svm_problem_x_set(prob, x_matrix)
def __init__(self,y,x): assert len(y) == len(x) self.prob = prob = svmc.new_svm_problem() self.size = size = len(y) self.y_array = y_array = svmc.new_double(size) for i in range(size): svmc.double_setitem(y_array,i,y[i]) self.x_matrix = x_matrix = svmc.svm_node_matrix(size) self.data = [] self.maxlen = 0; for i in range(size): data = _convert_to_svm_node_array(x[i]) self.data.append(data); svmc.svm_node_matrix_set(x_matrix,i,data) if type(x[i]) == dict: if (len(x[i]) > 0): self.maxlen = max(self.maxlen,max(x[i].keys())) else: self.maxlen = max(self.maxlen,len(x[i])) svmc.svm_problem_l_set(prob,size) svmc.svm_problem_y_set(prob,y_array) svmc.svm_problem_x_set(prob,x_matrix)
def cross_validation(prob, param, fold):
if param.gamma == 0:
param.gamma = 1.0 / prob.maxlen
dblarr = svmc.new_double(prob.size)
svmc.svm_cross_validation(prob.prob, param.param, fold, dblarr)
ret = _double_array_to_list(dblarr, prob.size)
svmc.delete_double(dblarr)
return ret
def _double_array(seq):
size = len(seq)
array = svmc.new_double(size)
i = 0
for item in seq:
svmc.double_setitem(array, i, item)
i = i + 1
return array
def cross_validation(prob, param, fold): if param.gamma == 0: param.gamma = 1.0/prob.maxlen dblarr = svmc.new_double(prob.size) svmc.svm_cross_validation(prob.prob, param.param, fold, dblarr) ret = _double_array_to_list(dblarr, prob.size) svmc.delete_double(dblarr) return ret
def _double_array(seq): size = len(seq) array = svmc.new_double(size) i = 0 for item in seq: svmc.double_setitem(array,i,item) i = i + 1 return array
def predict_values_raw(self, x):
#convert x into svm_node, allocate a double array for return
n = self.nr_class * (self.nr_class - 1) // 2
data = _convert_to_svm_node_array(x)
dblarr = svmc.new_double(n)
svmc.svm_predict_values(self.model, data, dblarr)
ret = _double_array_to_list(dblarr, n)
svmc.delete_double(dblarr)
svmc.svm_node_array_destroy(data)
return ret
def predict_values_raw(self,x): #convert x into svm_node, allocate a double array for return n = self.nr_class*(self.nr_class-1)//2 data = _convert_to_svm_node_array(x) dblarr = svmc.new_double(n) svmc.svm_predict_values(self.model, data, dblarr) ret = _double_array_to_list(dblarr, n) svmc.delete_double(dblarr) svmc.svm_node_array_destroy(data) return ret
def __init__(self, prob, param): if param is None: raise NotImplementedError, "svm_varma can't load a svm_model from a file (yet)" self.prob = prob if param.gamma == 0: param.gamma = 1.0/prob.maxlen msg = svmc.svm_check_parameter(prob.prob,param.param) if msg: raise ValueError, msg self.model = svmc.svm_train(prob.prob,param.param) #setup some classwide variables self.nr_class = nr_class = svmc.svm_get_nr_class(self.model) self.svm_type = svmc.svm_get_svm_type(self.model) #create labels(classes) intarr = svmc.new_int(self.nr_class) svmc.svm_get_labels(self.model,intarr) self.labels = labels = _int_array_to_list(intarr, self.nr_class) svmc.delete_int(intarr) if len(labels) != 2: raise NotImplementedError, "svm_varma doesn't handle problems with more than 2 labels (yet)" #check if valid probability model self.probability = svmc.svm_check_probability_model(self.model) model = self.model self.obj = svmc.svm_varma_get_obj(model, 0) self.rho = svmc.svm_varma_get_rho(model, 0) self.nSV = svmc.svm_varma_get_nSV(model, 0) self.total_sv = svmc.svm_varma_get_total_sv(model) # XXX: this only work for 2-class problems size = self.total_sv doublearr = svmc.new_double(size) svmc.svm_varma_get_sv_coef(model, 0, doublearr) self.sv_coef = _double_array_to_list(doublearr, size) svmc.delete_double(doublearr) # XXX: only works with PRECOMPUTED kernels size = self.total_sv doublearr = svmc.new_double(size) svmc.svm_varma_get_SV(model, doublearr) self.SV = _double_array_to_list(doublearr, size) svmc.delete_double(doublearr)
def predict_probability(self, x):
# c code will do nothing on wrong type, so we have to check ourself
if self.svm_type == NU_SVR or self.svm_type == EPSILON_SVR:
raise TypeError, "call get_svr_probability or get_svr_pdf for probability output of regression"
elif self.svm_type == ONE_CLASS:
raise TypeError, "probability not supported yet for one-class problem"
# only C_SVC,NU_SVC goes in
if not self.probability:
raise TypeError, "model does not support probabiliy estimates"
# convert x into svm_node, alloc a double array to receive probabilities
data = _convert_to_svm_node_array(x)
dblarr = svmc.new_double(self.nr_class)
pred = svmc.svm_predict_probability(self.model, data, dblarr)
pv = _double_array_to_list(dblarr, self.nr_class)
svmc.delete_double(dblarr)
svmc.svm_node_array_destroy(data)
p = {}
for i in range(len(self.labels)):
p[self.labels[i]] = pv[i]
return pred, p
def predict_probability(self,x):
#c code will do nothing on wrong type, so we have to check ourself
if self.svm_type == NU_SVR or self.svm_type == EPSILON_SVR:
raise TypeError, "call get_svr_probability or get_svr_pdf for probability output of regression"
elif self.svm_type == ONE_CLASS:
raise TypeError, "probability not supported yet for one-class problem"
#only C_SVC,NU_SVC goes in
if not self.probability:
raise TypeError, "model does not support probabiliy estimates"
#convert x into svm_node, alloc a double array to receive probabilities
data = _convert_to_svm_node_array(x)
dblarr = svmc.new_double(self.nr_class)
pred = svmc.svm_predict_probability(self.model, data, dblarr)
pv = _double_array_to_list(dblarr, self.nr_class)
svmc.delete_double(dblarr)
svmc.svm_node_array_destroy(data)
p = {}
for i in range(len(self.labels)):
p[self.labels[i]] = pv[i]
return pred, p
def get_model_params(model):
"""
Extract the alpha and b parameters from the SVM model.
returns (alpha, b)
"""
rho = svmc.svm_get_model_rho(model.model)
n = svmc.svm_get_model_num_coefs(model.model)
coefs_dblarr = svmc.new_double(n)
perm_intarr = svmc.new_int(n)
try:
svmc.svm_get_model_coefs(model.model,coefs_dblarr)
svmc.svm_get_model_perm(model.model,perm_intarr)
coefs = np.zeros(n,dtype=float)
perm = np.zeros(n,dtype=int)
for i in range(n):
coefs[i] = svmc.double_getitem(coefs_dblarr,i)
perm[i] = svmc.int_getitem(perm_intarr,i)
finally:
svmc.delete_double(coefs_dblarr)
svmc.delete_int(perm_intarr)
return (coefs, perm, rho)
def get_model_params(model):
"""
Extract the alpha and b parameters from the SVM model.
returns (alpha, b)
"""
rho = svmc.svm_get_model_rho(model.model)
n = svmc.svm_get_model_num_coefs(model.model)
coefs_dblarr = svmc.new_double(n)
perm_intarr = svmc.new_int(n)
try:
svmc.svm_get_model_coefs(model.model,coefs_dblarr)
svmc.svm_get_model_perm(model.model,perm_intarr)
coefs = np.zeros(n,dtype=float)
perm = np.zeros(n,dtype=int)
for i in range(n):
coefs[i] = svmc.double_getitem(coefs_dblarr,i)
perm[i] = svmc.int_getitem(perm_intarr,i)
finally:
svmc.delete_double(coefs_dblarr)
svmc.delete_int(perm_intarr)
return (coefs, perm, rho)