Performance different in R vs. Python!

#1

I’m experiencing different results in R and Python when I’m making a really basic model on the iris data set.
94% vs. 0.97% holdout accuracy, but for a more complex project I’m getting almost 15 percentage points difference!

For the test on the iris data set, the default parameters are the same and I’m using the same seeds.
I know the RNG is different between the languages, but is this really the reason or are there other reasons causing these differences?

#2

Yes, R uses its own RNG.

#3

But 15 percentage points is really a lot!
What is the RNG used for internally?

#4

We use RNG from the standard C++ library. Keep in mind that reproducibility is only guaranteed if you use the same language, OS, and compiler.