在Python中二进制化float64 Pandas Dataframe

Question

I've got a Panda DF with various columns (each indicating the frequency of a word in a corpus). Each row corresponds to a document and each is of type float64.

for example:

word1 word2 word3
0.0   0.3   1.0
0.1   0.0   0.5
etc

I want to Binarize this and instead of the frequency end up with a boolean (0s and 1s DF) that indicates the existence of a word

so the above example would be transformed to :

word1 word2 word3
0      1     1
1      0     1
etc

I looked at get_dummies(), but the output was not the expected.

Answer 1

Casting to boolean will result in True for anything that is not zero — and False for any zero entry. If you then cast to integer, you get ones and zeroes.

import io
import pandas as pd

data = io.StringIO('''\
word1 word2 word3
0.0   0.3   1.0
0.1   0.0   0.5
''')
df = pd.read_csv(data, delim_whitespace=True)

res = df.astype(bool).astype(int)
print(res)

Output:

   word1  word2  word3
0      0      1      1
1      1      0      1

Answer 2

I would have answered as @Alberto Garcia-Raboso answered but here is an alternative that is very quick and leverages the same idea.

Use np.where

pd.DataFrame(np.where(df, 1, 0), df.index, df.columns)

---

Timing

Answer 3

Code:

import numpy as np
import pandas as pd

""" create some test-data """
random_data = np.random.random([3, 3])
random_data[0,0] = 0.0
random_data[1,2] = 0.0

df = pd.DataFrame(random_data,
     columns=['A', 'B', 'C'], index=['first', 'second', 'third'])

print(df)

""" binarize """
threshold = lambda x: x > 0
df_ = df.apply(threshold).astype(int)

print(df_)

Output:

A         B         C
first   0.000000  0.610263  0.301024
second  0.728070  0.229802  0.000000
third   0.243811  0.335131  0.863908
A  B  C
first   0  1  1
second  1  1  0
third   1  1  1

Remarks:

• get_dummies() analyze each unique value per column and introduces new columns (for each unique value) to mark if this value is active
• = if column A has 20 unique values, 20 new columns are added, where exactly one column is true, the others are false

Answer 4

Found an alternative way using Pandas Indexing.

This can be simply done by

df[df>0] = 1

simple as that!