The goal is to filter a DataFrame on a dynamic number of columns with their respective individual values. To achieve this, I've created a filter mask from a dictionary which I should be able to use each time.

However this filter mask becomes a string and therefore provides a 'KeyError'. Some example of how my logic works.

import pandas as pd

# Create a list of dictionaries with the data for each row
data = [{'col1': 1, 'col2': 'a', 'col3': True, 'col4': 1.0},
        {'col1': 2, 'col2': 'b', 'col3': False, 'col4': 2.0},
        {'col1': 1, 'col2': 'c', 'col3': True, 'col4': 3.0},
        {'col1': 2, 'col2': 'd', 'col3': False, 'col4': 4.0},
        {'col1': 1, 'col2': 'e', 'col3': True, 'col4': 5.0}]
df = pd.DataFrame(data)

filter_dict = {'col1': 1, 'col3': True,}

def create_filter_query_for_df(filter_dict):
    query = ""
    for i, (column, values) in enumerate(filter_dict.items()):
        if i > 0:
            query  = " & "
        if isinstance(values,float) or isinstance(values,int):
            query  = f"(data['{column}'] == {values})"
        else:
            query  = f"(data['{column}'] == '{values}')"
    return query

df[create_filter_query_for_df(filter_dict)]

Result is:

KeyError: "(data['col1'] == 1) & (data['col3'] == True)"

The issue is that the create_filter_query_for_df() will return a string and it should be boolean variable. If you would make the mask as following:

mask1 = "(data['col1'] == 1) & (data['col3'] == True)" # the same error is returned;

# However if you format as below, it provides a success
mask2 = (data['col1'] == 1) & (data['col3'] == True)

The type of mask1 will be str. The type of mask2 will be boolean.

However, I can't use bool(mask1) because then I can't use it anymore as filter condition. I'm quite stuck so need some help here.

Apologies if I took a completely wrong approach in trying to get to the filter, it seemed quite a suitable solution to me.

Thanks in advance!

CodePudding user response：

The result of filtering based on mask2 is as follows:

mask2 = (df['col1'] == 1) & (df['col3'] == True)
df[mask2]

   col1 col2  col3  col4
0     1    a  True   1.0
2     1    c  True   3.0
4     1    e  True   5.0

To reach the same result with a string, we can use df.query like so:

df.query('(col1 == 1) & (col3 == True)')

   col1 col2  col3  col4
0     1    a  True   1.0
2     1    c  True   3.0
4     1    e  True   5.0

Note that the required syntax is actually a bit different. So, let's simplify your function to end up with the string that we need:

def create_filter_query_for_df(filter_dict):
    list_pairs = [f"({col} == {val})" for col, val in filter_dict.items()]
    query = ' & '.join(list_pairs)
    
    # '(col1 == 1) & (col3 == True)'
    
    return query

df.query(create_filter_query_for_df(filter_dict))

   col1 col2  col3  col4
0     1    a  True   1.0
2     1    c  True   3.0
4     1    e  True   5.0

ALternative approach

Incidentially, if you are only using the & operator, another way to approach this problem could be as follows:

• Use a list comprehension to create two pd.Series and use them as input for pd.concat with axis parameter set to 1.
• Chain df.all with axis parameter again set to 1 to evaluate if all values for each row in the resulting temporary df equal True).
• The result is a single pd.Series with booleans that we can use to filter the df.

my_mask = (pd.concat([df[k].eq(v) for k, v in filter_dict.items()], 
                     axis=1)
           .all(axis=1))

df[my_mask]

   col1 col2  col3  col4
0     1    a  True   1.0
2     1    c  True   3.0
4     1    e  True   5.0

Of course, this approach may not be ideal (or: function at all) if your actual requirements are a bit more complex.