def etri_infer_parameters(
learning_alts, criteria, pt, affectations, nprofiles, model, weights=None, lbda=None, profiles=None
):
categories = [(i + 1) for i in range(nprofiles + 1)]
categories_rank = {}
for i, cat in enumerate(categories):
categories_rank[cat] = i + 1
infile = glpk.create_input_file(
learning_alts, criteria, pt, categories, categories_rank, affectations, weights, lbda, profiles
)
print "GLPK file:", infile.name
(status, output) = glpk.solve(infile.name, model)
if status:
sys.exit("gklp returned status %d" % status)
infile.close()
return glpk.parse_output(output, learning_alts, criteria)
def main(argv=None):
if argv is None:
argv = sys.argv
# Parsing of the options
(in_dir, out_dir) = parse_cmdline(argv)
if error_list:
return error_list
check_input_files(in_dir)
if error_list:
create_error_file(out_dir, error_list)
return error_list
xmcda_input = parse_xmcda_files(in_dir)
if xmcda_input == None or error_list:
create_error_file(out_dir, error_list)
return error_list
alt_id = xmcda_input[0]
crit_id = xmcda_input[1]
pt = xmcda_input[2]
cat_id = xmcda_input[3]
cat_rank = xmcda_input[4]
assign = xmcda_input[5]
pref_dir = xmcda_input[6]
# Kludge: invert the categories rank... Beurk
invert_cat_rank(cat_rank)
cat_id_sorted = get_sorted_categories(cat_rank)
# Name of the profile alternatives
palts_id = [ "b%d" % (i+1) for i in range(len(cat_id)-1) ]
log("GLPK INPUT:")
log('alt ids : %s' % alt_id)
log('crit ids : %s' % crit_id)
log('perfs : %s' % pt)
log('categories : %s' % cat_id)
log('affect : %s' % assign)
log('pref_dir : %s' % pref_dir)
check_input_parameters(alt_id, crit_id, pt, cat_id, assign)
if error_list:
create_error_file(out_dir, error_list)
return error_list
fixed_params = get_fixed_parameters(in_dir, alt_id, crit_id, pt, cat_id, cat_rank, pref_dir)
fixed_weights = fixed_params[0]
fixed_lambda = fixed_params[1]
fixed_cat_profiles = fixed_params[2]
fixed_pt_profiles = fixed_params[3]
log('fixed_weights : %s' % fixed_weights)
log('fixed_lambda : %s' % fixed_lambda)
log('fixed_cat_profiles : %s' % fixed_cat_profiles)
log('fixed_pt_profiles : %s' % fixed_pt_profiles)
try:
(gmin, gmax) = get_min_max(pt)
if fixed_pt_profiles:
(gmin2, gmax2) = get_min_max(fixed_pt_profiles)
for c, val in gmin.iteritems():
if gmin[c] > gmin2[c]:
gmin[c] = gmin2[c]
if gmax[c] < gmax2[c]:
gmax[c] = gmax2[c]
except:
error_list.append("Impossible to get min max values in pt")
create_error_file(out_dir, error_list)
return error_list
try:
normalize(pt, gmin, gmax, pref_dir)
except:
error_list.append("Impossible to convert performance table")
create_error_file(out_dir, error_list)
return error_list
if fixed_pt_profiles:
try:
normalize(fixed_pt_profiles, gmin, gmax, pref_dir)
fixed_profiles = get_sorted_profiles(fixed_cat_profiles, cat_rank, fixed_pt_profiles)
except:
error_list.append("Impossible to convert profile performance table")
create_error_file(out_dir, error_list)
return error_list
else:
fixed_profiles = None
if fixed_weights:
glpk_model = "inf_etri_bm_fixed_weights.mod"
elif fixed_pt_profiles:
glpk_model = "inf_etri_bm_fixed_profiles.mod"
else:
glpk_model = "inf_etri_bm.mod"
log("GLPK model used: %s" % glpk_model)
try:
input_file = glpk.create_input_file(alt_id, crit_id, pt, cat_id, cat_rank, assign, fixed_weights, fixed_lambda, fixed_profiles)
except:
error_list.append("Impossible to create glpk input file")
if error_list:
create_error_file(out_dir, error_list)
return error_list
(status, output) = glpk.solve(glpk_model, input_file.name)
if status:
error_list.append("GLPK returned status %d" % status);
input_file.close()
create_error_file(out_dir, error_list)
return error_list
input_file.close()
out_params = glpk.parse_output(output, alt_id, crit_id)
if out_params == None:
error_list.append("Error parsing output parameters")
create_error_file(out_dir, error_list)
return error_list
weights = out_params[0]
profiles = out_params[1]
for profile in profiles:
denormalize(profile, gmin, gmax, pref_dir)
lbda = out_params[2]
compat = out_params[3]
log("GLPK OUTPUT:")
log("weights : %s" % weights)
log("profiles : %s" % profiles)
log("lambda : %s" % lbda)
log("compat : %s" % compat)
if not weights:
error_list.append("Impossible to get weights from the solver")
if not profiles:
error_list.append("Impossible to get profiles from the solver")
if not lbda:
error_list.append("Impossible to get lambda from the solver")
if not compat:
error_list.append("Impossible to get compatible alternatives from the solver")
if error_list:
create_error_file(out_dir, error_list)
return error_list
out_weights = xmcda.format_criteria_weights(weights)
out_catprof = xmcda.format_category_profiles(profiles, palts_id, cat_id_sorted)
out_refalts = xmcda.format_pt_reference_alternatives(profiles, palts_id, crit_id)
out_lambda = xmcda.format_lambda(lbda)
out_compat = xmcda.format_format_compatible_alternatives(compat, alt_id)
create_output_files(out_dir, out_weights, out_catprof, out_refalts, out_lambda, out_compat)
logs = [ "Execution ok" ]
create_log_file(out_dir, logs)
# Categories
categories = [(i+1) for i in range(len(profiles)+1)]
categories_rank = {}
for i, cat in enumerate(categories):
categories_rank[cat] = i+1
learning_alts = pt.keys()[0:50]
# ELECTRE TRI model
model = etri.electre_tri(pt, profiles, weights, lbda)
pessimist = model.pessimist()
optimist = model.optimist()
#debug.print_performance_table_with_assignements(pt, alternatives, criteria, pessimist)
infile = glpk.create_input_file(learning_alts, criteria, pt, categories, categories_rank, pessimist)
(status, output) = glpk.solve_normalized(infile.name)
infile.close()
if status:
sys.exit("gklp returned status %d" % status)
(iweights, iprofiles, ilbda, icompat) = glpk.parse_output(output, learning_alts, criteria)
if iweights == None:
sys.exit("Invalid weights");
if iprofiles == None:
sys.exit("Invalid profiles")
if ilbda == None:
sys.exit("Invalid lambda")
if icompat == None:
sys.exit("Invalid compat");