Problems using groupby for word count then loop using python

Question

Word Count and cumulative sum

I have a data set up to 1.5 millions rows. This data set is a time series is a year format as shown below. I am trying to count the strings per year in a cumulative format. Example below:

lodgement_year                trademark_text
  1906                          PEPS
  1906  BILE BEANS FOR BILIOUSNESS B
  1906                     ZAM-BUK Z

  lodgement_year
  1906    {PEPS BILE BEANS FOR BILIOUSNESS B ZAM-BUK Z Z...

  1907    {WHS CHERUB BLACK & WHITE SOUTHERN CROSS HISTO...

As a initial task I grouped the strings then applied a loop in all year using the code that was posted in this forum by xxx . While the loop works the following message appears straight after:

    The code :

    d = df_merge.groupby('lodgement_year')['trademark_text'].apply(lambda x: "{%s}" % ' '.join(x))


  for name in d.index:
            data = d.loc[name]
            ngram_vectorizer =    CountVectorizer(analyzer='word',tokenizer=word_tokenize, ngram_range=(1, 1), min_df=1)              
            X = ngram_vectorizer.fit_transform(data.split('\n'))
            vocab = list(ngram_vectorizer.get_feature_names())
            counts = X.sum(axis=0).A1
            freq_distribution = Counter(dict(zip(vocab, counts)))
            print (name, freq_distribution.most_common(10))

The error message:

Traceback (most recent call last):

File "/Users/PycharmProjects/Slice_Time_Series", line 65, in X = ngram_vectorizer.fit_transform(data.split('\n'))

File "/Users/anaconda3/lib/python3.6/site-packages/pandas/core/generic.py", line 3081, in getattr return object.getattribute(self, name) AttributeError: 'Series' object has no attribute 'split'

The output that works before the error:

 1906 [('.', 24), ("'s", 22), ('star', 18), ('&', 15), ('kodak', 12), ('co', 9), ('the', 9), ('brand', 8), ('express', 8), ('anchor', 6)]
 1907 [('&', 11), ("'s", 11), ('brand', 11), ('pinnacle', 7), ('vaseline', 7), ('the', 6), ('.', 5), ('co.', 5), ('kepler', 5), ('lucas', 5)]

Any help will be greatly appreciated. As a next task Im trying to create a series which is a sum of 1906 then 1906 plus 1907 then 1906+1907+1908 I have no idea what to do yet, any guidance would be also great.

Ian

Answer 1

You can first group your DF by the lodgement variable, and then iterate over name-group pairs, do another groupby, and finally join the results.

import pandas as pd
from collections import Counter
df = pd.DataFrame({'lodg':[1,2,1,3,1,2,2,3,1,1],
               'text':['x y z','y y','x y','z x','y x','y y z','x z','x x','x x','y z']})
grouped = df.groupby('lodg')
joined = []
for name, group in grouped:
    texts = " ".join(group['text'])
    groupCounts = Counter(texts.split(" ")).items()
    joined.append([name, [texts], groupCounts])
groupedJoined = pd.DataFrame(joined, columns = ['lodg','texts','textCounts'])

Result:

In [16]: groupedJoined
Out[16]: 
   lodg                    texts                textCounts
0     1  [x y z x y y x x x y z]  [(y, 4), (x, 5), (z, 2)]
1     2          [y y y y z x z]  [(y, 4), (x, 1), (z, 2)]
2     3                [z x x x]          [(x, 3), (z, 1)]

After the explicit solution, it took a little to come up with the 1-liner lambda solution.

Two Lambdas:

df.groupby('lodg')['text'].apply(lambda x: "%s" % ' '.join(x)).apply(lambda x: Counter(x.split(" ")).items())

Single Lambda:

df.groupby('lodg')['text'].apply(lambda x: Counter((' '.join(x)).split(" ")).items())

Both produce the same result:

Out[62]: 
lodg
1    [(y, 4), (x, 5), (z, 2)]
2    [(y, 4), (x, 1), (z, 2)]
3            [(x, 3), (z, 1)]
Name: text, dtype: object

Now, if you don't want to deal with the unnamed column (actually called index), name the result res, and do res.reset_index() to get this:

In [68]: res.reset_index()
Out[68]: 
   lodg                      text
0     1  [(y, 4), (x, 5), (z, 2)]
1     2  [(y, 4), (x, 1), (z, 2)]
2     3          [(x, 3), (z, 1)]

This should be much more straightforward to work with.