KNN regressor | Python Fiddle

import math
from sklearn.neighbors import NearestNeighbors
import numpy as np
import pandas as pd
k =4
df = pd.read_csv(r'C:\Users\bkalluri\Downloads\data_2.csv')
predicted_list = []
for index, row in df.iterrows() :
    list1 = []
    a = row.latitude
    b= row.longitude
    list1.append(a)
    list1.append(b)
    point1 = np.array(list1).reshape((1,2))
    array1 = pd.DataFrame(df.loc[df['close_date']<row.close_date,['latitude','longitude','close_price']]).as_matrix()
    '''
        Proceed only if an entry is found in the array. Array would be without any sample if there are no neighbours to any home.
        This happens for the home with minimum value of close date and without any duplicates.
    '''
    if array1.shape[0] is not 0:
        '''
        Checking the number of neighbours. If the neighbour count is < 4, 
        then K is tightened to neighbours found. This is to accomodate the problem constraint.
        '''
        if array1.shape[0] <= k :
            k = array1.shape[0]
        nbrs = NearestNeighbors(n_neighbors=k, algorithm='ball_tree').fit(array1[:,[0,1]])
        distances, indices = nbrs.kneighbors(point1)
        distances = distances.reshape(k,)
        indices = indices.reshape(k,)
        prices = pd.Series(array1[:,2])
        df2 = pd.DataFrame({'indices':indices,'distances':distances},index=indices)
        df2['ratio'] = np.exp(-df2['distances'])
        df2['weight'] = df2['ratio']/df2['ratio'].sum(axis=0)
        df2['prices'] = prices
        df2['estimate'] = df2['weight'] * df2['prices']
        predicted_value = df2['estimate'].sum(axis=0)
        predicted_list.append(predicted_value)
        #print(predicted_value)
        '''
        If the Home doesn't have any neighbour, the predicted value is considered to be none.

'''
    else :
        predicted_list.append(None)
#print(len(df.index))
#print(len(predicted_list))
df['predicted_price'] = np.array(predicted_list)
print(df)

Python Fiddle

Python Cloud IDE