Slicing a multi-index pandas dataframe with a large list

najeem :

I have a large dataframe with multi-index. I wanted to slice this dataframe using a fairly large list. Below is a sample code. It is taking almost 10 seconds for this operation.

import pandas as pd
import numpy as np

df = pd.DataFrame(
    {
        "x": np.repeat(np.arange(10000), 50),
        "y": np.repeat(np.arange(50), 10000),
        "val": np.random.rand(50*10000)
    }
).set_index(["x", "y"])

large_list = range(5000,10000)

slice = df.loc[(large_list, slice(None)),:] # Takes 10 seconds on my machine

As a comparison, if I write this dataframe to an hdf file and read it with a where condition same as my slicing operation, it takes only 1.5 seconds!

df.to_hdf("sample.hdf", key="df", append=True)
df1 = pd.read_hdf("sample.hdf", "df", where='x in large_list')

Is there a faster way to slice in memory?

Andy L. :

If your intention is slicing multiindex by an arbitrary list, using query will be much faster

Create an arbitrary list from 5000 to 10000

np.random.seed(0)
large_list =  np.random.choice(list(range(5000, 10000)), 5000, replace=False)

In [2245]: large_list
Out[2245]: array([5398, 8833, 9836, ..., 6653, 7607, 7732])

x = df.query('x in @large_list')

Compare result

In [2246]: y = df.loc[(large_list, slice(None)),:]
In [2249]: np.allclose(x, y)
Out[2249]: True