对R中的序数逻辑回归进行交叉验证（使用rpy2）

Question

I'm trying to create a predictive model in Python, comparing several different regression models through cross-validation. In order to fit an ordinal logistic model ( MASS.polr ), I've had to interface with R through rpy2 as follows:

from rpy2.robjects.packages import importr
import rpy2.robjects as ro

df = pd.DataFrame()
df = df.append(pd.DataFrame({"y":25,"X":7},index=[0]))
df = df.append(pd.DataFrame({"y":50,"X":22},index=[0]))
df = df.append(pd.DataFrame({"y":25,"X":15},index=[0]))
df = df.append(pd.DataFrame({"y":75,"X":27},index=[0]))
df = df.append(pd.DataFrame({"y":25,"X":12},index=[0]))
df = df.append(pd.DataFrame({"y":25,"X":13},index=[0]))

# Loads R packages. 
base = importr('base')
mass = importr('MASS')

# Converts df to an R dataframe. 
from rpy2.robjects import pandas2ri
pandas2ri.activate()
ro.globalenv["rdf"] = pandas2ri.py2ri(df) 

# Makes R recognise y as a factor. 
ro.r("""rdf$y <- as.factor(rdf$y)""")

# Fits regression. 
formula = "y ~ X"    
ordlog = mass.polr(formula, data=base.as_symbol("rdf"))
ro.globalenv["ordlog"] = ordlog
print(base.summary(ordlog))

So far, I have mainly been comparing my models using sklearn.cross_validation.test_train_split and sklearn.metrics.accuracy_score , yielding a number from 0 to 1 which represents the accuracy of the training-set model in predicting the test-set values.

How might I replicate this test using rpy2 and MASS.polr ?

Answer 1

通过使用rms.lrm重新拟合模型最终解决了问题，该模型提供了validate()函数（在此示例之后进行解释）。