def generate_grid_search(hash_model_selection, kernel):
c1 = hash_model_selection['soft_margin1']
c2 = hash_model_selection['soft_margin2']
step_c = hash_model_selection['step_soft_margin']
if c1 >= c2:
raise Exception(
'ERROR',
'run_svm.generate_grid_search: The limit points of soft_margin parameter are not properly configured'
)
if step_c <= 0:
raise Exception(
'ERROR',
'run_svm.generate_grid_search: The incremental amount for soft_margin scanning is not properly configured'
)
grid_soft_margin = frange(c1, c2, step_c)
if len(grid_soft_margin) <= 1:
raise Exception(
'ERROR',
'run_svm.generate_grid_search: The list of soft_margin parameter has one element or is empty'
)
if kernel == 'linear':
return grid_soft_margin, []
g1 = hash_model_selection['gamma1']
g2 = hash_model_selection['gamma2']
step_g = hash_model_selection['step_gamma']
if g1 >= g2:
raise Exception(
'ERROR',
'run_svm.generate_grid_search: The limit points of gamma parameter are not properly configured'
)
if step_g <= 0:
raise Exception(
'ERROR',
'run_svm.generate_grid_search: The incremental amount for gamma scanning is not properly configured'
)
grid_gamma = frange(g1, g2, step_g)
if len(grid_gamma) <= 1:
raise Exception(
'ERROR',
'run_svm.generate_grid_search: The list of gamma parameter has one element or is empty'
)
return grid_soft_margin, grid_gamma
def generate_grid_search(hash_model_selection,kernel):
c1 = hash_model_selection['soft_margin1']
c2 = hash_model_selection['soft_margin2']
step_c = hash_model_selection['step_soft_margin']
if c1 >= c2:
raise Exception('ERROR','run_svm.generate_grid_search: The limit points of soft_margin parameter are not properly configured')
if step_c <= 0:
raise Exception('ERROR','run_svm.generate_grid_search: The incremental amount for soft_margin scanning is not properly configured')
grid_soft_margin = frange(c1,c2,step_c)
if len(grid_soft_margin) <= 1:
raise Exception('ERROR','run_svm.generate_grid_search: The list of soft_margin parameter has one element or is empty')
if kernel == 'linear':
return grid_soft_margin,[]
g1 = hash_model_selection['gamma1']
g2 = hash_model_selection['gamma2']
step_g = hash_model_selection['step_gamma']
if g1 >= g2:
raise Exception('ERROR','run_svm.generate_grid_search: The limit points of gamma parameter are not properly configured')
if step_g <= 0:
raise Exception('ERROR','run_svm.generate_grid_search: The incremental amount for gamma scanning is not properly configured')
grid_gamma = frange(g1,g2,step_g)
if len(grid_gamma) <= 1:
raise Exception('ERROR','run_svm.generate_grid_search: The list of gamma parameter has one element or is empty')
return grid_soft_margin,grid_gamma
def calculate_roc_values(self,l_scores,l_labels): """ Bibliography: ROC Graphs: Notes and Practical Considerations for Researchers Tom Fawcett ([email protected]) HP Laboratories, MS 1143, 1501 Page Mill Road, Palo Alto, CA 94304 """ if len(l_scores) <= 1: return [] l_scores_sort = sorted(l_scores) mini = l_scores_sort[0] maxi = l_scores_sort[-1] increment = max(l_scores_sort[1]-l_scores_sort[0],min_increment) l_margin_thres = frange(mini,maxi,increment) l_margin_thres.append(1) num_cpu = 2 if cpu_count() > 1: num_cpu = cpu_count() pool = Pool(num_cpu-1) result = pool.map_async(get_confusion_matrix, map(lambda x: (l_scores,l_labels,x) , l_margin_thres)) result.wait() l_results = result.get() l_roc_fpr = map(lambda x: self.__get_FPR(x['FP'],x['N']),l_results) l_roc_tpr = map(lambda x: self.__get_TPR(x['TP'],x['P']),l_results) return sorted(zip(l_roc_fpr,l_roc_tpr))
