eleanora eleanora - 2 months ago 7
Python Question

How to recode and count efficiently

I have a large csv with three strings per row in this form:

a,c,d
c,a,e
f,g,f
a,c,b
c,a,d
b,f,s
c,a,c


I read in the first two columns recode the strings to integers and then remove duplicates counting how many copies of each row there were as follows:

import pandas as pd
df = pd.read_csv("test.csv", usecols=[0,1], prefix="ID_", header=None)
letters = set(df.values.flat)
df.replace(to_replace=letters, value=range(len(letters)), inplace=True)
df1 = df.groupby(['ID_0', 'ID_1']).size().rename('count').reset_index()
print df1


This gives:

ID_0 ID_1 count
0 0 1 2
1 1 0 3
2 2 4 1
3 4 3 1


which is exactly what I need.

However as my data is large I would like to make two improvements.


  • How can I do the groupby and then recode instead of the other way round? The problem is that I can't do df1[['ID_0','ID_0']].replace(to_replace=letters, value=range(len(letters)), inplace = True). This gives the error "A value is trying to be set on a copy of a slice from a DataFrame"

  • How can I avoid creating df1? That is do the whole thing inplace.


Answer

I like to use sklearn.preprocessing.LabelEncoder to do the letter to digit conversion:

from sklearn.preprocessing import LabelEncoder

# Perform the groupby (before converting letters to digits).
df = df.groupby(['ID_0', 'ID_1']).size().rename('count').reset_index()

# Initialize the LabelEncoder.
le = LabelEncoder()
le.fit(df[['ID_0', 'ID_1']].values.flat)

# Convert to digits.
df[['ID_0', 'ID_1']] = df[['ID_0', 'ID_1']].apply(le.transform)

The resulting output:

   ID_0  ID_1  count
0     0     2      2
1     1     3      1
2     2     0      3
3     3     4      1

If you want to convert back to letters at a later point in time, you can use le.inverse_transform:

df[['ID_0', 'ID_1']] = df[['ID_0', 'ID_1']].apply(le.inverse_transform)

Which maps back as expected:

  ID_0 ID_1  count
0    a    c      2
1    b    f      1
2    c    a      3
3    f    g      1

If you just want to know which digit corresponds to which letter, you can look at the le.classes_ attribute. This will give you an array of letters, which is indexed by the digit it encodes to:

le.classes_ 

['a' 'b' 'c' 'f' 'g']

For a more visual representation, you can cast as a Series:

pd.Series(le.classes_)

0    a
1    b
2    c
3    f
4    g

Timings

Using a larger version of the sample data and the following setup:

df2 = pd.concat([df]*10**5, ignore_index=True)

def root(df):
    df = df.groupby(['ID_0', 'ID_1']).size().rename('count').reset_index()
    le = LabelEncoder()
    le.fit(df[['ID_0', 'ID_1']].values.flat)
    df[['ID_0', 'ID_1']] = df[['ID_0', 'ID_1']].apply(le.transform)
    return df

def pir2(df):
    unq = np.unique(df)
    mapping = pd.Series(np.arange(unq.size), unq)
    return df.stack().map(mapping).unstack() \
      .groupby(df.columns.tolist()).size().reset_index(name='count')

I get the following timings:

%timeit root(df2)
10 loops, best of 3: 101 ms per loop

%timeit pir2(df2)
1 loops, best of 3: 1.69 s per loop
Comments