uproot4 to uproot5 TTree to DataFrame arrays slower

[masterfelu]

Hello.

I run a script to load a TTree as a DataFrame with 5 million rows and 3 columns, and measure the time taken. Here's an MWE for ipython:

%timeit uproot.open('tree.root')['time'].arrays(['year','month','day'],library='pd')

For uproot 5.3.7 with pandas 2.2.2, 19.2 s ± 177 ms per loop
For uproot 4.1.9 with pandas 1.3.5, 1.61 s ± 9.27 ms per loop

The value reported is mean ± std. dev. of 7 runs, 1 loop each. We can see uproot5 is 4 times slower than uproot4. The discrepancy increases further when more columns are loaded, leading to more than ten minutes of time for what previously took seconds.

I just want to know if there has been any major change that can cause such a reduction of load time for large TTree to DataFrame. Also, if any more checks should be done before I reach a conclusion, that is super helpful.

Thanks a lot for your time.

[vvsagar]

It might also be important to mention the differences in the pandas versions in order to reproduce the results?

[JoyYTZhou]

I've also noticed the same issue with uproot.open('someNANOAOD12.root:Events').arrays(lib='ak'). This file has 680 entries and a size of 4.053MB. I am timing only the operation of getting the arrays for the tree object, and it takes 20.8 seconds.
I never had such issues before with much larger files with some older version of uproot (I think it was still 5 but a particular release). I am now using version 5.4.1. I am not sure if it's an issue with uproot or awkward. I am leaning toward it being an uproot error since I've had my awkward version fixed in virtual env.