import pandas as pd
import sklearn.datasets as datasets
import tensorflow as tf

data_boston = datasets.load_boston()
df = pd.DataFrame(data_boston.data, columns=data_boston.feature_names).drop(['ZN', 'CHAS', 'RAD'], axis=1)
X, Y = df.values, data_boston.target.reshape(-1, 1)
from sklearn.model_selection import train_test_split
X_train,X_test,Y_train,Y_test=train_test_split(X,Y,test_size=0.2)

num_featurs=X.shape[1]

sess=tf.Session()

k=4
batch_size=20

x_train=tf.placeholder(shape=[None,num_featurs],dtype=tf.float32)
x_test=tf.placeholder(shape=[None,num_featurs],dtype=tf.float32)
y_train=tf.placeholder(shape=[None,1],dtype=tf.float32)
y_test=tf.placeholder(shape=[None,1],dtype=tf.float32)


distance=tf.reduce_sum(tf.abs(x_train-tf.expand_dims(x_test,1)),axis=2)
# 返回一个距离矩阵（使用了tf的广播计算原理）
# 这里用的是l1范数。如果想用l2范数，只需要把tf.abs换成tf.quare，

top_k_xvals, top_k_indices = tf.nn.top_k(-distance, k=k)
top_k_yvals = tf.gather(y_train, top_k_indices)
prediction=tf.reduce_mean(top_k_yvals,axis=1)

top_k_xvals, top_k_indices = tf.nn.top_k(-distance, k=k)
x_sums = tf.expand_dims(tf.reduce_sum(top_k_xvals, 1), 1)
x_sums_repeated = tf.matmul(x_sums, tf.ones((1, k), dtype=tf.float32))
x_val_weights = tf.expand_dims(tf.div(top_k_xvals, x_sums_repeated), 1)
# 主要为了用距离做权重

top_k_yvals = tf.gather(y_train, top_k_indices)
prediction = tf.squeeze(tf.matmul(x_val_weights, top_k_yvals), squeeze_dims=[1])
# mse = tf.div(tf.square(prediction - y_test), batch_size)

Y_test_predict=sess.run(prediction,feed_dict={x_train:X_train,x_test:X_test,y_train:Y_train})

from sklearn.metrics import r2_score
r2_score(Y_test,Y_test_predict)

# 附上一个sklearn的实现
from sklearn.neighbors import KNeighborsRegressor
knr=KNeighborsRegressor(n_neighbors=k)
knr.fit(X_train,Y_train)
Y_test_predict1=knr.predict(X_test)
r2_score(Y_test,Y_test_predict1)

import tensorflow as tf
import numpy as np
from tensorflow.examples.tutorials.mnist import input_data

mnist = input_data.read_data_sets('MNIST_data/', one_hot=True)

train_size = 1000
test_size = 100
rand_train_indices = np.random.choice(len(mnist.train.images), train_size)
rand_test_indices = np.random.choice(len(mnist.test.images), test_size)
X_train = mnist.train.images[rand_train_indices]
Y_train = mnist.train.labels[rand_train_indices]
X_test = mnist.test.images[rand_test_indices]
Y_test = mnist.test.labels[rand_test_indices]
num_featurs = X_train.shape[1]

sess=tf.Session()
k=4
x_train=tf.placeholder(shape=[None,num_featurs],dtype=tf.float32)
x_test=tf.placeholder(shape=[None,num_featurs],dtype=tf.float32)
y_train=tf.placeholder(shape=[None,10],dtype=tf.float32)
y_test=tf.placeholder(shape=[None,10],dtype=tf.float32)

distance=tf.reduce_sum(tf.abs(x_train-tf.expand_dims(x_test,1)),axis=2)

top_k_xvals, top_k_indices = tf.nn.top_k(-distance, k=k)
top_k_yvals = tf.gather(y_train, top_k_indices)

prediction=tf.reduce_mean(top_k_yvals,axis=1)

Y_test_predict=sess.run(prediction,feed_dict={x_train:X_train,x_test:X_test,y_train:Y_train})

# 也可以用tf框架做：tf.argmax(prediction,dimension=1)
predict = [i.argmax() for i in Y_test_predict]
true = [i.argmax() for i in Y_test]
correct_count, total_count = 0, 0
for i in range(len(predict)):
    if predict[i] == true[i]:
        correct_count += 1
    total_count += 1

correct_count/total_count