How to speed up pandas row filtering by string matching?

Question

I often need to filter pandas dataframe df by df[df['col_name']=='string_value'], and I want to speed up the row selction operation, is there a quick way to do that ?

For example,

In [1]: df = mul_df(3000,2000,3).reset_index()

In [2]: timeit df[df['STK_ID']=='A0003']
1 loops, best of 3: 1.52 s per loop

Can 1.52s be shorten ?

Note:

mul_df() is function to create multilevel dataframe:

>>> mul_df(4,2,3)
                 COL000  COL001  COL002
STK_ID RPT_Date                        
A0000  B000      0.6399  0.0062  1.0022
       B001     -0.2881 -2.0604  1.2481
A0001  B000      0.7070 -0.9539 -0.5268
       B001      0.8860 -0.5367 -2.4492
A0002  B000     -2.4738  0.9529 -0.9789
       B001      0.1392 -1.0931 -0.2077
A0003  B000     -1.1377  0.5455 -0.2290
       B001      1.0083  0.2746 -0.3934

Below is the code of mul_df():

import itertools
import numpy as np
import pandas as pd

def mul_df(level1_rownum, level2_rownum, col_num, data_ty='float32'):
    ''' create multilevel dataframe, for example: mul_df(4,2,6)'''

    index_name = ['STK_ID','RPT_Date']
    col_name = ['COL'+str(x).zfill(3) for x in range(col_num)]

    first_level_dt = [['A'+str(x).zfill(4)]*level2_rownum for x in range(level1_rownum)]
    first_level_dt = list(itertools.chain(*first_level_dt)) #flatten the list
    second_level_dt = ['B'+str(x).zfill(3) for x in range(level2_rownum)]*level1_rownum

    dt = pd.DataFrame(np.random.randn(level1_rownum*level2_rownum, col_num), columns=col_name, dtype = data_ty)
    dt[index_name[0]] = first_level_dt
    dt[index_name[1]] = second_level_dt

    rst = dt.set_index(index_name, drop=True, inplace=False)
    return rst

Answer 1

I have long wanted to add binary search indexes to DataFrame objects. You can take the DIY approach of sorting by the column and doing this yourself:

In [11]: df = df.sort('STK_ID') # skip this if you're sure it's sorted

In [12]: df['STK_ID'].searchsorted('A0003', 'left')
Out[12]: 6000

In [13]: df['STK_ID'].searchsorted('A0003', 'right')
Out[13]: 8000

In [14]: timeit df[6000:8000]
10000 loops, best of 3: 134 µs per loop

This is fast because it always retrieves views and does not copy any data.

Answer 2

Somewhat surprisingly, working with the .values array instead of the Series is much faster for me:

>>> time df = mul_df(3000, 2000, 3).reset_index()
CPU times: user 5.96 s, sys: 0.81 s, total: 6.78 s
Wall time: 6.78 s
>>> timeit df[df["STK_ID"] == "A0003"]
1 loops, best of 3: 841 ms per loop
>>> timeit df[df["STK_ID"].values == "A0003"]
1 loops, best of 3: 210 ms per loop