utility.py

import gurobipy as gp
import numpy as np
import matplotlib.pyplot as plt
import random
from random import randint
from numba import jit, njit

def load_points(file):
    f = open(file, 'r')
    points = []

    for i,line in enumerate(f.readlines()):
        #ritorna il count nella prima variabile e il value nella seconda
        nums = line.split(" ")
        coords = []
        for num in nums:
            if num != '' and num != '\n':
                coords.append(float(num))
        if len(coords)!=0:
            coords = np.array(coords)
            points.append(coords)

    points = np.array(points)
    f.close()
    return points

@jit(nopython=True)#modalità di compilazione ad alte performance
def calc_centroids(sol, points, K):
    dim = np.shape(points)[1]
    centroids = np.zeros((K,dim))
    num_elems = np.zeros(K)
    for elem,cluster in enumerate(sol):
        cluster = int(cluster)
        elem = int(elem)
        centroids[cluster] = centroids[cluster] + points[elem]
        num_elems[cluster] = num_elems[cluster] + 1

    for i in range(K):
        centroids[i] = centroids[i] / num_elems[i]
    
    return centroids

@jit(nopython=True)
def squared_inner_distance(sol, points, K):
    dim = np.shape(points)[1]
    centroids = np.zeros((K,dim))
    num_elems = np.zeros(K)
    for elem,cluster in enumerate(sol):
        cluster = int(cluster)
        elem = int(elem)
        centroids[cluster] = centroids[cluster] + points[elem]
        num_elems[cluster] = num_elems[cluster] + 1

    for i in range(K):
        centroids[i] = centroids[i] / num_elems[i]
    
    tot_distance = 0

    for elem,cluster in enumerate(sol):
        cluster = int(cluster)
        elem = int(elem)
        tot_distance = np.sum((centroids[cluster]-points[elem])**2) + tot_distance

    return tot_distance

def printR2sol(points, sol, K):
    N = len(sol)
    colors = []
    for i in range(K):
        colors.append('#%06X' % randint(0, 0xFFFFFF))

    clusters = []

    for i in range(K):
        cluster = []
        for j in range(N):
            if(sol[j]==i):
                cluster.append(points[j].copy())
        cluster = np.array(cluster)
        clusters.append(cluster)


    for i in range(K):
        to_draw = clusters[i]
        plt.scatter(to_draw[:,0], to_draw[:,1], color = colors[i])
    plt.show()

def create_initial_sol(points, K):
    N = points.shape[0]
    dimension = points.shape[1]
    centroids = np.zeros((K,dimension))
    clusters = np.zeros(N)
    choices = np.arange(N)

    for i in range(K):
        choice = np.random.randint(len(choices))
        centroids[i] = points[choice].copy() #centroids_list[i]
        choices = np.delete(choices, choice)
   
        for i in range(N):
            dist = -1
            centroid = -1
            for c in range(K):
                dist_c = np.linalg.norm(centroids[c]-points[i])
                if(dist_c < dist or dist == -1):
                    centroid = c
                    dist = dist_c
            clusters[i] = int(centroid)

    return clusters