在多行中使用str.contains

Question

I have a dataframe with five rows that looks like this:

index  col1   col2  col3   col4   col5
1      word1  None  word1  None   None
2      None   word1 word2  None   None
3      None   None  None   word2  word2
4      word1  word2 None   None   None

I'm trying to find all rows that contain both strings in any combination of columns---in this case, rows 2 and 4. Normally I would use the str.contains method to filter by string:

df[df['col1'].str.contains('word1 | word2'), case=False)

But this only gives me A) results for one column, and B) a True if the column has one word. I intuitively tried df[df[['col1', 'col2', 'col3', 'col4', 'col5']].str.contains('word1' & 'word2'), case=False) but .str.contains doesn't work on DataFrame objects.

Is there a way to do this without resorting to a for loop?

Answer 1

Using any

s1=df.apply(lambda x : x.str.contains(r'word1')).any(1)
s2=df.apply(lambda x : x.str.contains(r'word2')).any(1)
df[s1&s2]
Out[452]: 
        col1   col2   col3  col4  col5
index                                 
2       None  word1  word2  None  None
4      word1  word2   None  None  None

Answer 2

If there is only 2 words you are looking for, You could use np.isin and any to check if each row in the underlying numpy array contains both the elements, using a separate isin for each word:

df[np.isin(df.values, 'word1').any(1) & np.isin(df.values, 'word2').any(1)]

   index   col1   col2   col3  col4  col5
1      2   None  word1  word2  None  None
3      4  word1  word2   None  None  None

Or, following the same logic but borrowing a bit from @coldspeed's answer:

words = ['word1','word2']

df[np.logical_and.reduce([np.isin(df.values, w).any(1) for w in words])]

   index   col1   col2   col3  col4  col5
1      2   None  word1  word2  None  None
3      4  word1  word2   None  None  None

Answer 3

Assuming you want only the rows with both word1 and word2 somewhere, you will need to stack , groupby index, and search inside an apply .

words = ['word1', 'word2']
df[df.stack().groupby(level=0).apply(
    lambda x: all(x.str.contains(w, case=False).any() for w in words))]

print(df)
        col1   col2   col3  col4  col5
index                                 
2       None  word1  word2  None  None  # word1=>col2, word2=>col3
4      word1  word2   None  None  None  # word1=>col1, word2=>col2

---

Another alternative would be using np.logical_and.reduce :

v = df.stack()
m = pd.Series(
        np.logical_and.reduce([
           v.str.contains(w, case=False).groupby(level=0).transform('any') 
           for w in words]),
        index=v.index)
df = df[m.unstack().all(1)]

print(df)
        col1   col2   col3  col4  col5
index                                 
2       None  word1  word2  None  None
4      word1  word2   None  None  None