R, Python, rpy2: "Error: cannot allocate vector of size xxx Mb"

user_2315976 0
Viewed 839

I work on Python 3.7 with R 3.5.1. When I call my function from the R script from Python it works in principle, however in the end it crashes with output:

rpy2.rinterface.RRuntimeError: Error: cannot allocate vector of size 1006.0 Mb

I have 16GB of RAM and I inspected during the long processing the fluctuations of memory usage which maxed at less than 5GB.

To make things even more certain of what's going on I traced memory usage with 

gc() 
gcinfo(TRUE)

which dumps garbage and tracks when R does so too by itself.

Additionally, in R I get the following, indicating 16k MB which is all my default RAM:

> memory.limit()
[1] 16244

Finally, when I run the R script straight from R it works fine and no memory issues are encountered.

Is it a formidable bug of rpy2 or something else that could be managed right now? If anyone has any suggestions they are most welcome, I will try.

OS: Windows 10, 64bit

rpy2 memory r python
1 Answers

From this thread :

This seems to be caused by hard-to-reconcile differences between...R vectors and Python arrays.

I can suggest doing the following as a workaround:

  • communicate the vector data via a mutual accessible place for both R and Python (e.g. database, csv file in a common folder etc.)
  • open a ticket on the rpy2 page (or continue the thread linked)
  • break up your data (if possible) to the maximum size the you can transfer via rpy2 and loop over it until you cover your whole data set (maybe a small alteration on both sides can achieve this - depending on your code)

  • consider a type of conversion that is offered in the thread, such as:

    from rpy2.robjects import conversion df_R = conversion.converter.py2ri(df_pandas)

Moderator note: this could be a comment, but I am unable to post comments

user_109436370
Related