Bdtjtk

I'm trying to implement Particle swarm optimization on support vector machine (SVM) to optimizing cost and gamma parameters (Radial basis function) to improve the accuracy.The SVM program is taking data from mysql db and is working fine.Now I need a PSO code to optimize the cost and gamma parameters and use it in svm program.
following are the svm and pso codes. Svm works fine but the PSO code does not work.
the code is written in python.
Please help..

SVM code:

import MySQLdb

import cgi

import cgitb

import pandas as pd 

import numpy as np 

from sklearn.cross_validation import train_test_split

from sklearn import preprocessing, cross_validation, svm 

from sklearn.svm import SVR

import mysql.connector as sql

import pandas as pd

cgitb.enable()

print 'Content-type: text/htmlrnr'

form = cgi.FieldStorage()

e1= form.getvalue('EARNING_PER_SHARE', '')

e2 = form.getvalue('CASH_INVESTMENT', '')

e3 = form.getvalue('CURRENT_LIABILITY', '')

e4 = form.getvalue('TOTAL_REVENUE', '')

e5 = form.getvalue('GROSS_PROFIT', '')

db = MySQLdb.connect(host="127.0.0.1", db="cisco", user="root", passwd="")

cursor = db.cursor()

cursor.execute("""

INSERT INTO table1 (EARNING_PER_SHARE, CASH_INVESTMENT, CURRENT_LIABILITY, 

TOTAL_REVENUE,GROSS_PROFIT) 

VALUES (%s, %s, %s, %s, %s)

""", (e1, e2, e3, e4,e5))

db.commit()

db.close()

db_connection = sql.connect(host='127.0.0.1', database='cisco', user='root', 

password='')

db_cursor = db_connection.cursor()

db_cursor.execute('SELECT * FROM table1')

table_rows = db_cursor.fetchall()

df = pd.DataFrame(table_rows)

np = df.as_matrix()

X = df.drop([4], 1)

np1 = X.as_matrix()

y = df[4]

np2 = y.as_matrix()

x_train, x_test, y_train, y_test = train_test_split(np1, np2, test_size=0.3)

clf = svm.SVR()

clf.fit(np1, np2)

confidence = clf.score(np1, np2)

for k in ['rbf']:

clf = svm.SVR(kernel=k, C=100, gamma=0.0001)

clf.fit(np1, np2)

confidence = clf.score(np1, np2)

print(k,confidence)

a = clf.predict(np1)

print ('npredicted values')

print (a)

print ('nreal values')

print (np2)

PSO code:

def fitness_function(a,x,np2,np1,C,gamma,):    

C = x[0]

gamma = x[1]

clf = svm.SVR(kernel=k, C=10, gamma=0.0001,swarmsize=50)

clf.fit(np1, np2)

confidence = clf.score(np1, np2)

print(k,confidence)

mse = sqrt(mean_squared_error(np2, a))

return mse

lb = [10, 0.0001]

ub = [1000,0.1]

xopt, fopt = pso(fitness_function, lb, ub)

I want to use RMSE as the fitness function but I think fitness function has to include the parameters that is to be optimized. But, in SVR, the objective function is too complicated to write in python.