我对使用GPU的经验很少，在网上查看后，我想到了这一点：
<xsl:stylesheet xmlns:xsl="http://www.w3.org/1999/XSL/Transform"
xmlns:xs="http://www.w3.org/2001/XMLSchema"
exclude-result-prefixes="#all"
version="3.0">
<xsl:mode on-no-match="shallow-copy"/>
<xsl:strip-space elements="*"/>
<xsl:output indent="yes"/>
<xsl:template match="sounds | sounds//*">
<xsl:where-populated>
<xsl:next-match/>
</xsl:where-populated>
</xsl:template>
</xsl:stylesheet>
import pandas as pd
from haversine import haversine
import numpy as np
from pandas import ExcelWriter
import numba as nb
np.set_printoptions(precision=20)
path = 'distance.xlsx'
df = pd.read_excel(path)
df = df.assign(Dist=pd.Series(np.zeros(27055)).values);
#df = df.assign(Facility=pd.Series(np.zeros(27055)).values);
df = df.assign(Facility=pd.Series(np.zeros((27055,),dtype='float,float')).values);
df["Facility_city"] = ""
#idx = np.asarray(df.loc[df["lat1"] != '.'].ix[:,0].index)
#temp1 = 1e10
#j = 0
idx = np.asarray(df[(df['lat1']!='.') & (df['state']== df['state'][0])].index)
@nb.jit(nopython=True)
def f(df):
temp1 = 1e10
j = 0
for i in range(0, len(df)):
if df['state'][i+1] != df['state'][i]:
idx = np.asarray(df[(df['lat1']!='.') & (df['state']== df['state'][i+1])].index)
#while (df.iloc[idx[j]]['state'] == df.iloc[i]['state']):
while (j!=len(idx)):
p1 = (df.iloc[idx[j]]['lat1'],df.iloc[idx[j]]['long1'])
p2 = (df.iloc[i]['lat2'],df.iloc[i]['long2'])
df.Dist.iloc[i] = min(temp1,haversine(p1, p2, miles=True))
if df.Dist.iloc[i] < temp1:
#df.Facility.iloc[i] = idx[j]
df.Facility.iloc[i] = (p1[0],p1[1])
df.Facility_city.iloc[i] = df.city.iloc[idx[j]]
temp1 = df.Dist.iloc[i]
j+=1
j = 0
temp1 = 1e10
return df
if __name__ == "__main__":
df = f(df)
writer = ExcelWriter('Results.xlsx')
df.to_excel(writer,'Sheet1')
writer.save()
运行近3万次。因此，我倾向于使用GPU（通过for loop），我正努力为代码配置相同的代码。
执行此操作时，出现错误。
Floydhub我该如何规避这个问题？我认为numba是参考文献中最好的框架，但其他任何框架（例如PyCUDA）也可以。