通過 rpy2 在 Python 中通過基於 R 的 glmmTMB 擬合 Logistic AR1

Question

我正在嘗試在 Python 中擬合具有串行自相關或 AR1 錯誤的邏輯回歸。 不幸的是， statsmodels還沒有走到這一步。 但是，R-package glmmTMB有。 我很接近（似乎）通過rpy2中的 rpy2 來解決這個問題，但我被卡住了。

My R skills (and the error: RRuntimeError: Error in na.fail.default(as.ts(x)): missing values in object ) suggest a need to modify the na.action argument to function glmmTMB (in the package glmmTMB ) ，但在 R 中這不是必需的——我的例子適合那里並且很高興。 所以，我懷疑 Python-to-R 鏈中的其他東西是錯誤的。

我的錢花在了變量time和group到formula中的ar部分所必需的因子表征上。 有人看到我能做什么嗎？ 我沒有正確創建因子變量嗎？

from rpy2.robjects.packages import importr
from rpy2.robjects import pandas2ri 
from rpy2 import robjects as ro
import pandas as pd

# Make some data.  
data = {'Success': [10, 20, 30, 11, 14, 16, 18, 29, 17, 19], 
        'Failure': [12, 25, 61, 8, 22, 21, 10, 16, 15, 19]}
df = pd.DataFrame(data)

# Allow rpy2 to talk to pandas.  
pandas2ri.activate() 

# Bring in some R stuff.  
base = importr('base')
utils = importr('utils')
stats = importr('stats')
glmmTMB = importr('glmmTMB')

# Modify the dataframe for use with glmmTMB.  
df['time'] = range(1, len(df) + 1)
df['group'] = 1

# Convert df to an R dataframe.  
r_df = pandas2ri.py2ri(df)

# Add in group as a factor to R dataframe r_df.
col_group = ro.vectors.FactorVector(r_df.rx2('group'))
assessor_col_index = df.columns.get_loc('group')   # python 
r_df[assessor_col_index] = col_group

# Add in time as a factor to R dataframe r_df.  
col_time = ro.vectors.FactorVector(r_df.rx2('time'))
assessor_col_index = df.columns.get_loc('time')   # python 
r_df[assessor_col_index] = col_time

# Take a look at stuff.  Looks correct.  
print(r_df)
print(col_group)
print(col_time) 

# This works.  Normal outcome.  
m0 = ro.r.glmmTMB(formula=ro.r('Success ~ 1'), data=r_df)
# print(m0[1][0])

# This works.  Logistic outcome.  
m1 = ro.r.glmmTMB(formula=ro.r('cbind(Success, Failure) ~ 1'), family=ro.r('binomial'), data=r_df)
# print(m1[1][0])

# This does not work.  AR1 with factors, per documentation.  
# m2 = ro.r.glmmTMB(formula=ro.r("cbind(Success, Failure) ~ 1 + ar(time + 0 | group)"), family=ro.r('binomial'), data=r_df)
# m2[1][0]


# Added for clarity, 2021-03-08:  Other variations using na_action argument.  

# m3  = ro.r.glmmTMB(formula=ro.r("cbind(Success, Failure) ~ 1 + ar(time + 0 | group)"), family=ro.r('binomial'), data=r_df, na_action=na_omit)            
# name 'na_omit' is not defined.

# m4  = ro.r.glmmTMB(formula=ro.r("cbind(Success, Failure) ~ 1 + ar(time + 0 | group)"), family=ro.r('binomial'), data=r_df, na_action=ro.r(na_omit))     
# name 'na_omit' is not defined. 

# m5  = ro.r.glmmTMB(formula=ro.r("cbind(Success, Failure) ~ 1 + ar(time + 0 | group)"), family=ro.r('binomial'), data=r_df, na_action='na_omit')        
# Error in na.fail.default(as.ts(x)): missing values in object 

# m6  = ro.r.glmmTMB(formula=ro.r("cbind(Success, Failure) ~ 1 + ar(time + 0 | group)"), family=ro.r('binomial'), data=r_df, na_action=ro.r('na_omit'))  
# Error in (function(expr, envir=parent.frame(), enclos=if(is.list(envir) || object 'na_omit' not found)))
 
# m7  = ro.r.glmmTMB(formula=ro.r("cbind(Success, Failure) ~ 1 + ar(time + 0 | group)"), family=ro.r('binomial'), data=r_df, na_action=na.omit)          
# name 'na' is not defined 

# m8  = ro.r.glmmTMB(formula=ro.r("cbind(Success, Failure) ~ 1 + ar(time + 0 | group)"), family=ro.r('binomial'), data=r_df, na_action=ro.r(na.omit))      
# name 'na' is not defined

# m9  = ro.r.glmmTMB(formula=ro.r("cbind(Success, Failure) ~ 1 + ar(time + 0 | group)"), family=ro.r('binomial'), data=r_df, na_action='na.omit')        
# Error in na.fail.default(as.ts(x)): missing values in object

# m10 = ro.r.glmmTMB(formula=ro.r("cbind(Success, Failure) ~ 1 + ar(time + 0 | group)"), family=ro.r('binomial'), data=r_df, na_action=ro.r('na.omit'))   
# Error in na.fail.default(as.ts(x)): missing values in object 

Answer 1

如果命名參數na.action在 function glmmTMB()的簽名中，您應該可以這樣做

m = ro.r.glmmTMB(<blah>, na_action=<your value>)

文檔在這里： https://rpy2.github.io/doc/v3.4.x/html/robjects_functions.html?highlight=function#