Reverse lookup of a sub-string in a dict of lists-CodePudding

I am struggling to find a "pythonic" way to make this logic work:

I have a dataframe of payment transactions with a column of strings( "beneficiary"):

index	beneficiary
12	REWE SAGT DANKE. ...
13	NaN
14	OBI BAU- U. HEIMWER//BERLIN/DE / OBI SAGT DANKE
15	NETFLIX INTERNATIONAL B.V.

I need to create another column in my dataframe which would be the category that every line belongs to.

index	beneficiary	category
12	REWE SAGT DANKE. ...	Groceries
14	OBI BAU- U. HEIMWER//BERLIN/DE / OBI SAGT DANKE	Groceries
15	NETFLIX INTERNATIONAL B.V.	Entertainment

I am thinking to make a dictionary like this below and somehow reverse look-up the sub-string values from the categories dict with the column above:

categories = {"Groceries": ["EDEKA", "REWE", "OBI"],
            "Entertainment": ["NETFLIX"]}

The logic is: if sub-string "REWE" is in df['beneficiary'] then df['category'] equals the key of the dict element where the sub-string is.

I am open to other mapping logic.

CodePudding user response：

You can use a regex, for this you need to rework your dictionary to have the matched strings as key and categories as values:

categories = {"Groceries": ["EDEKA", "REWE", "OBI"],
              "Entertainment": ["NETFLIX"]}

cat_sub = {v:k for k,l in categories.items() for v in l}
regex = r'(%s)' % '|'.join(fr'\b{c}\b' for c in cat_sub)
# regex looks like this: (\bEDEKA\b|\bREWE\b|\bOBI\b|\bNETFLIX\b)

df['category'] = df['beneficiary'].str.extract(regex, expand=False).map(cat_sub)

NB. I used word boundaries (\b) to ensure matching full words, use regex = r'(%s)' % '|'.join(cat_sub) if you don't want this behaviour

output:

   index                                       beneficiary       category
0      12                             REWE SAGT DANKE. ...      Groceries
1      13                                              NaN            NaN
2      14  OBI BAU- U. HEIMWER//BERLIN/DE / OBI SAGT DANKE      Groceries
3      15                       NETFLIX INTERNATIONAL B.V.  Entertainment

NB. if needed to drop the NaNs, use dropna

CodePudding user response：

Try reversing your dictionary to map each word to a category and applying this to each word in "beneficiary":

word_cat = {w: k for k,v in categories.items() for w in v}
df["category"] = df["beneficiary"].str.split(expand=True).apply(lambda x: x.map(word_cat)).bfill(axis=1).iloc[:,0]
df = df.dropna()

>>> df
   index                                    beneficiary       category
0     12                               REWE SAGT DANKE.      Groceries
2     14  OBI BAU-U. HEIMWER//BERLIN/DE /OBI SAGT DANKE      Groceries
3     15                     NETFLIX INTERNATIONAL B.V.  Entertainment