Applied Spatial Statistics（二）统计推断：排列测试
本笔记本演示了如何执行假设检验和 p 值计算的排列检验。
- 相关系数
- Moran’s I
相关性
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
from scipy.stats import *url = 'https://raw.githubusercontent.com/mwaskom/seaborn-data/master/mpg.csv'df = pd.read_csv(url)
df = df.dropna()
让我们抽取一个非常小的样本，只有 10 个点。
small_sample_car = df.sample(10,random_state=222)
small_sample_car
| mpg | cylinders | displacement | horsepower | weight | acceleration | model_year | origin | name | |
|---|---|---|---|---|---|---|---|---|---|
| 165 | 20.0 | 8 | 262.0 | 110.0 | 3221 | 13.5 | 75 | usa | chevrolet monza 2+2 |
| 380 | 36.0 | 4 | 120.0 | 88.0 | 2160 | 14.5 | 82 | japan | nissan stanza xe |
| 16 | 18.0 | 6 | 199.0 | 97.0 | 2774 | 15.5 | 70 | usa | amc hornet |
| 261 | 18.1 | 6 | 258.0 | 120.0 | 3410 | 15.1 | 78 | usa | amc concord d/l |
| 19 | 26.0 | 4 | 97.0 | 46.0 | 1835 | 20.5 | 70 | europe | volkswagen 1131 deluxe sedan |
| 99 | 18.0 | 6 | 232.0 | 100.0 | 2945 | 16.0 | 73 | usa | amc hornet |
| 206 | 26.5 | 4 | 140.0 | 72.0 | 2565 | 13.6 | 76 | usa | ford pinto |
| 86 | 14.0 | 8 | 304.0 | 150.0 | 3672 | 11.5 | 73 | usa | amc matador |
| 267 | 27.5 | 4 | 134.0 | 95.0 | 2560 | 14.2 | 78 | japan | toyota corona |
| 232 | 16.0 | 8 | 351.0 | 149.0 | 4335 | 14.5 | 77 | usa | ford thunderbird |
plt.scatter(small_sample_car.horsepower, small_sample_car.acceleration)
plt.xlabel("HP")
plt.ylabel("Acceleration")
Text(0, 0.5, 'Acceleration')
基于此小样本的 Pearson 相关性，请注意此处的分析 p 值为 ~0.03
pearson_rslt = pearsonr(small_sample_car.horsepower, small_sample_car.acceleration)print("Pearson:",pearson_rslt)
Pearson: PearsonRResult(statistic=-0.6832264558183371, pvalue=0.02941732498278316)
np.random.permutation(small_sample_car.acceleration)
array([20.5, 13.6, 11.5, 15.5, 14.5, 13.5, 14.5, 15.1, 16. , 14.2])
定义相关性的排列测试函数：
def permutation_cor(sample_car):null_cor_list = []for i in range(10000):#permutate one variable, while hold the other constant.subsample_Y = np.random.permutation(sample_car.acceleration)#compute correlationsample_cor = pearsonr(sample_car.horsepower, subsample_Y)[0]null_cor_list.append(sample_cor)return null_cor_list这给了我们零假设下的相关系数。
null_dist = permutation_cor(small_sample_car)
plt.hist(null_dist)plt.vlines(pearson_rslt[0],ymin=0,ymax=2000,color="red")
<matplotlib.collections.LineCollection at 0x15ff5d6d0>
让我们计算一下 unll 下的相关性（在幅度上）比样本相关性大多少倍。
这为我们提供了一个模拟 p 值。
np.sum(np.abs(np.array(null_dist)) >= np.abs(pearson_rslt[0]))
323
import numpy as np
p_sim = np.sum(np.abs(np.array(null_dist)) >= np.abs(pearson_rslt[0]))/10000print("The simulated p-value (2-tail): ", p_sim)print("The actual p-value (2-tail) : ", pearson_rslt[1])
The simulated p-value (2-tail): 0.0323
The actual p-value (2-tail) : 0.02941732498278316
Moran’s I
让我们看看伊利诺伊州 ALAND 的空间自相关。
import geopandas as gpd
IL = gpd.read_file("https://raw.githubusercontent.com/Ziqi-Li/GIS5122/main/data/IL_acs/IL_acs.shp")
IL.columns
Index(['state', 'county', 'NAME_x', 'total_pop', 'pct_65_ove', 'pct_black','pct_hispan', 'males', 'females', 'pct_bach', 'GEOID_x', 'STATEFP','COUNTYFP', 'COUNTYNS', 'AFFGEOID', 'GEOID_y', 'NAME_y', 'LSAD','ALAND', 'AWATER', 'GEOID_int', 'geometry'],dtype='object')
IL.plot(column="ALAND",legend=True)
<Axes: >
使用 esda 包计算 Moran’s I 值。
from libpysal.weights.contiguity import Queenvalue = IL['ALAND'].values #Get the data we are interested.w = Queen.from_dataframe(IL) #Generate W matrix using Queen contiguityw.transform = 'r' #Row-standardization of the weight matrix
/var/folders/mp/9px298sd6vs8xccb_3sql0dr0000gp/T/ipykernel_45186/246861888.py:5: FutureWarning: `use_index` defaults to False but will default to True in future. Set True/False directly to control this behavior and silence this warningw = Queen.from_dataframe(IL) #Generate W matrix using Queen contiguity
from esda.moran import Moran
from splot.esda import plot_moranil_moran = Moran(value, w, permutations=1000, two_tailed=True)print("Moran's I:", il_moran.I)print("analytical p-value:",il_moran.p_norm)print("Simulated p-value:",il_moran.p_sim)
Moran's I: 0.16857257061968103
analytical p-value: 0.0035814725750113125
Simulated p-value: 0.004995004995004995
plot_moran(il_moran, zstandard=True, figsize=(10,4))
plt.show()
/Users/ziqili/anaconda3/lib/python3.11/site-packages/splot/_viz_esda_mpl.py:354: FutureWarning: `shade` is now deprecated in favor of `fill`; setting `fill=True`.
This will become an error in seaborn v0.14.0; please update your code.sbn.kdeplot(moran.sim, shade=shade, color=color, ax=ax, **kwargs)
#Define a permutation test function for Moran's I:
def permutation_moran(geodataframe):w = Queen.from_dataframe(IL) null_moran_list = []for i in range(1000):value = np.random.permutation(geodataframe['ALAND'].values) #Get the data we are interested.#compute moran's Imoran = Moran(value, w)null_moran_list.append(moran.I)return null_moran_list我们为每个县随机分配一个值，并计算每个地图的 Moran’s I。
import numpy as np
IL.plot(column=np.random.permutation(IL['ALAND'].values),legend=True)
<Axes: >
原假设下的 Moran’s I
null_moran = permutation_moran(IL)
/var/folders/mp/9px298sd6vs8xccb_3sql0dr0000gp/T/ipykernel_45186/1735371652.py:4: FutureWarning: `use_index` defaults to False but will default to True in future. Set True/False directly to control this behavior and silence this warningw = Queen.from_dataframe(IL)
plt.hist(null_moran)
plt.vlines(il_moran.I,ymin=0,ymax=300,color="red")
<matplotlib.collections.LineCollection at 0x2a57a3f10>
p_sim = np.sum(np.abs(np.array(null_moran)) >= np.abs(il_moran.I))/10000print("Analytical p-value:",il_moran.p_norm)
print("Simulated p-value from esda's permutation:",il_moran.p_sim)print("Simulated p-value from permutation test:",p_sim)
Analytical p-value: 0.0035814725750113125
Simulated p-value from esda's permutation: 0.004995004995004995
Simulated p-value from permutation test: 0.0005
转存中…(img-XgiyYpYB-1713438830006)]
p_sim = np.sum(np.abs(np.array(null_moran)) >= np.abs(il_moran.I))/10000print("Analytical p-value:",il_moran.p_norm)
print("Simulated p-value from esda's permutation:",il_moran.p_sim)print("Simulated p-value from permutation test:",p_sim)
Analytical p-value: 0.0035814725750113125
Simulated p-value from esda's permutation: 0.004995004995004995
Simulated p-value from permutation test: 0.0005
)
)
