HarmonyCore.py

# -*- coding: utf-8 -*-
"""
Created on Sat Jan 20 13:11:19 2018

@author: [TWkiwi] Maxwell Chen
"""

import numpy as np
class HarmonyCore(object):
    
    def __init__(self,harmony_obj):
        self.obj_func = harmony_obj
        self.hmm_matrix = list()
        matrix = []
        for limit in self.obj_func.up_down_limit:
            row = np.random.uniform(low=limit[0], high=limit[1], size=(1,self.obj_func.harmony_menmory_size))[0] 
            matrix.append(row)
        matrix = np.asarray(matrix).transpose().round(self.obj_func.weight_decimal)
        print(matrix)
        self.hmm_matrix = matrix
        
    def run(self):
        hmm_err_list = [0] * len(self.hmm_matrix)
        for m_i in range(len(self.hmm_matrix)):
            vetor_list = self.hmm_matrix[m_i]

            error = self.obj_func.fitness(vetor_list,self.obj_func.input_X,self.obj_func.input_Y)
            hmm_err_list[m_i] = error
            
        for itera in range(self.obj_func.iteration):
            vetor_list = [0] * self.obj_func.vector_size
            #while True:
            for i in range(self.obj_func.vector_size):
                if np.random.rand(1,)[0] < self.obj_func.hmcr_proba:
                    #new_vactor = 0.0
                    new_vactor = self.hmm_matrix[np.random.randint(self.obj_func.harmony_menmory_size, size=1)[0]][i]
                    if np.random.rand(1,)[0] < self.obj_func.par_proba:
                        if np.random.rand(1,)[0] < self.obj_func.adju_proba:
                            #new_vactor -= np.std(self.hmm_matrix[:][i]) * np.random.rand(1,)[0]
                            new_vactor -= (new_vactor - self.obj_func.up_down_limit[i][0]) * np.random.rand(1,)[0]
                        else:
                            #new_vactor += np.std(self.hmm_matrix[:][i]) * np.random.rand(1,)[0]
                            new_vactor += (self.obj_func.up_down_limit[i][0] - new_vactor) * np.random.rand(1,)[0]
                    vetor_list[i] = round(new_vactor, self.obj_func.weight_decimal)
                else: 
                    new_vactor = np.random.uniform(low=self.obj_func.up_down_limit[i][0], high=self.obj_func.up_down_limit[i][1], size=(1,))[0]
                    vetor_list[i] = round(new_vactor, self.obj_func.weight_decimal)
                '''
                if not vetor_list in self.hmm_matrix:
                    break
                '''
            #隨機選Sample
            if self.obj_func.sample_size > 0:
                random_idx = np.random.permutation(len(self.obj_func.input_X))[:self.obj_func.sample_size]
            else:
                random_idx = np.random.permutation(len(self.obj_func.input_X))
            error = self.obj_func.fitness(vetor_list,[self.obj_func.input_X[i] for i in random_idx],[self.obj_func.input_Y[i] for i in random_idx])
            overwrite_index = hmm_err_list.index(max(hmm_err_list))
            if hmm_err_list[overwrite_index] >= error:
                assert([round(i, self.obj_func.weight_decimal) for i in vetor_list] == vetor_list)
                vetor_list = [round(i, self.obj_func.weight_decimal) for i in vetor_list]
                if not vetor_list in self.hmm_matrix.tolist():
                    #print('HMM_UPDATEP_NEW_VECTOR:',vetor_list)
                    #print('HMCR:',self.obj_func.hmcr_proba,'PAR:',self.obj_func.par_proba,'HMM_UPDATE_NEW_ERROR:',error)
                    hmm_err_list[overwrite_index] = error
                    self.hmm_matrix[overwrite_index] = vetor_list
            #else:
            #    print(vetor_list,'worst than',self.hmm_matrix[overwrite_index],'because new error:',error,'higher than',hmm_err_list[overwrite_index])
        return self.hmm_matrix,hmm_err_list,hmm_err_list.index(min(hmm_err_list))