from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.neighbors import KNeighborsClassifier
import numpy as np

# 載入數(shù)據(jù)集
iris_dataset = load_iris()

# 數(shù)據(jù)劃分
X_train, X_test, y_train, y_test = train_test_split(iris_dataset['data'], iris_dataset['target'], random_state=0)

# 設(shè)置鄰居數(shù)
knn = KNeighborsClassifier(n_neighbors=1)

# 構(gòu)建基于訓(xùn)練集的模型
knn.fit(X_train, y_train)


# 一條測(cè)試數(shù)據(jù)
X_new = np.array([[5, 2.9, 1, 0.2]])

# 對(duì)X_new預(yù)測(cè)結(jié)果
prediction = knn.predict(X_new)
print("預(yù)測(cè)值%d" % prediction)

# 得出測(cè)試集X_test測(cè)試集的分?jǐn)?shù)
print("score:{:.2f}".format(knn.score(X_test,y_test)))

import csv
import random
import numpy as np
import operator


def openfile(filename):

  """
  打開(kāi)數(shù)據(jù)集，進(jìn)行數(shù)據(jù)處理
  :param filename: 數(shù)據(jù)集的路徑
  :return: 返回?cái)?shù)據(jù)集的數(shù)據(jù)，標(biāo)簽，以及標(biāo)簽名
  """

  with open(filename) as csv_file:
    data_file = csv.reader(csv_file)
    temp = next(data_file)

    # 數(shù)據(jù)集中數(shù)據(jù)的總數(shù)量
    n_samples = int(temp[0])

    # 數(shù)據(jù)集中特征值的種類個(gè)數(shù)
    n_features = int(temp[1])

    # 標(biāo)簽名
    target_names = np.array(temp[2:])

    # empty()函數(shù)構(gòu)造一個(gè)未初始化的矩陣，行數(shù)為數(shù)據(jù)集數(shù)量，列數(shù)為特征值的種類個(gè)數(shù)
    data = np.empty((n_samples, n_features))

    # empty()函數(shù)構(gòu)造一個(gè)未初始化的矩陣，行數(shù)為數(shù)據(jù)集數(shù)量，1列，數(shù)據(jù)格式為int
    target = np.empty((n_samples,), dtype=np.int)

    for i, j in enumerate(data_file):

      # 將數(shù)據(jù)集中的將數(shù)據(jù)轉(zhuǎn)化為矩陣，數(shù)據(jù)格式為float
      # 將數(shù)據(jù)中從第一列到倒數(shù)第二列中的數(shù)據(jù)保存在data中
      data[i] = np.asarray(j[:-1], dtype=np.float64)

      # 將數(shù)據(jù)集中的將數(shù)據(jù)轉(zhuǎn)化為矩陣，數(shù)據(jù)格式為int
      # 將數(shù)據(jù)集中倒數(shù)第一列中的數(shù)據(jù)保存在target中
      target[i] = np.asarray(j[-1], dtype=np.int)

  # 返回 數(shù)據(jù)，標(biāo)簽 和標(biāo)簽名
  return data, target, target_names


def random_number(data_size):
  """
  該函數(shù)使用shuffle()打亂一個(gè)包含從0到數(shù)據(jù)集大小的整數(shù)列表。因此每次運(yùn)行程序劃分不同，導(dǎo)致結(jié)果不同

  改進(jìn)：
  可使用random設(shè)置隨機(jī)種子，隨機(jī)一個(gè)包含從0到數(shù)據(jù)集大小的整數(shù)列表，保證每次的劃分結(jié)果相同。

  :param data_size: 數(shù)據(jù)集大小
  :return: 返回一個(gè)列表
  """

  number_set = []
  for i in range(data_size):
    number_set.append(i)

  random.shuffle(number_set)

  return number_set


def split_data_set(data_set, target_data, rate=0.25):
  """
  說(shuō)明：分割數(shù)據(jù)集，默認(rèn)數(shù)據(jù)集的25%是測(cè)試集

  :param data_set: 數(shù)據(jù)集
  :param target_data: 標(biāo)簽數(shù)據(jù)
  :param rate: 測(cè)試集所占的比率
  :return: 返回訓(xùn)練集數(shù)據(jù)、訓(xùn)練集標(biāo)簽、訓(xùn)練集數(shù)據(jù)、訓(xùn)練集標(biāo)簽
  """

  # 計(jì)算訓(xùn)練集的數(shù)據(jù)個(gè)數(shù)
  train_size = int((1-rate) * len(data_set))

  # 獲得數(shù)據(jù)
  data_index = random_number(len(data_set))

  # 分割數(shù)據(jù)集（X表示數(shù)據(jù)，y表示標(biāo)簽），以返回的index為下標(biāo)
  x_train = data_set[data_index[:train_size]]

  x_test = data_set[data_index[train_size:]]

  y_train = target_data[data_index[:train_size]]

  y_test = target_data[data_index[train_size:]]
  return x_train, x_test, y_train, y_test


def data_diatance(x_test, x_train):
  """
  :param x_test: 測(cè)試集
  :param x_train: 訓(xùn)練集
  :return: 返回計(jì)算的距離
  """

  # sqrt_x = np.linalg.norm(test-train) # 使用norm求二范數(shù)（距離）
  distances = np.sqrt(sum((x_test - x_train) ** 2))
  return distances


def knn(x_test, x_train, y_train, k):
  """
  :param x_test: 測(cè)試集數(shù)據(jù)
  :param x_train: 訓(xùn)練集數(shù)據(jù)
  :param y_train: 測(cè)試集標(biāo)簽
  :param k: 鄰居數(shù)
  :return: 返回一個(gè)列表包含預(yù)測(cè)結(jié)果
  """

  # 預(yù)測(cè)結(jié)果列表，用于存儲(chǔ)測(cè)試集預(yù)測(cè)出來(lái)的結(jié)果
  predict_result_set=[]

  # 訓(xùn)練集的長(zhǎng)度
  train_set_size = len(x_train)

  # 創(chuàng)建一個(gè)全零的矩陣，長(zhǎng)度為訓(xùn)練集的長(zhǎng)度
  distances = np.array(np.zeros(train_set_size))

  # 計(jì)算每一個(gè)測(cè)試集與每一個(gè)訓(xùn)練集的距離
  for i in x_test:
    for indx in range(train_set_size):

      # 計(jì)算數(shù)據(jù)之間的距離
      distances[indx] = data_diatance(i, x_train[indx])

    # 排序后的距離的下標(biāo)
    sorted_dist = np.argsort(distances)

    class_count = {}

    # 取出k個(gè)最短距離
    for i in range(k):

      # 獲得下標(biāo)所對(duì)應(yīng)的標(biāo)簽值
      sort_label = y_train[sorted_dist[i]]

      # 將標(biāo)簽存入字典之中并存入個(gè)數(shù)
      class_count[sort_label]=class_count.get(sort_label, 0) + 1

    # 對(duì)標(biāo)簽進(jìn)行排序
    sorted_class_count = sorted(class_count.items(), key=operator.itemgetter(1), reverse=True)

    # 將出現(xiàn)頻次最高的放入預(yù)測(cè)結(jié)果列表
    predict_result_set.append(sorted_class_count[0][0])

  # 返回預(yù)測(cè)結(jié)果列表
  return predict_result_set


def score(predict_result_set, y_test):
  """

  :param predict_result_set: 預(yù)測(cè)結(jié)果列表
  :param y_test: 測(cè)試集標(biāo)簽
  :return: 返回測(cè)試集精度
  """
  count = 0
  for i in range(0, len(predict_result_set)):
    if predict_result_set[i] == y_test[i]:
      count += 1

  score = count / len(predict_result_set)

  return score


if __name__ == "__main__":

  iris_dataset = openfile('iris.csv')
  # x_new = np.array([[5, 2.9, 1, 0.2]])
  x_train, x_test, y_train, y_test = split_data_set(iris_dataset[0], iris_dataset[1])
  result = knn(x_test,x_train, y_train, 6)
  print("原有標(biāo)簽:", y_test)

  # 為了方便對(duì)比查看，此處將預(yù)測(cè)結(jié)果轉(zhuǎn)化為array,可直接打印結(jié)果
  print("預(yù)測(cè)結(jié)果：", np.array(result))
  score = score(result, y_test)
  print("測(cè)試集的精度：%.2f" % score)