CODEWITHSUNDEEP
pythonpandasmachine-learning
Samira Kumar
Samira Kumar

Reputation: 521

Converting count vectorizer to tf-idf

So I've the following table with each row is a document and each column is words and no of occurrence of the words.

|doc|apple|banana|cat| 
|---|---|---|---| 
|1|2|0|0| 
|2|0|0|2| 
|3|0|2|0|

Is there any method to convert these count vectorized table to tf-idf vectorizer?

Edit: My solution for it. Let me know if this is correct.

def get_tfidf(df_tfidf):

total_docs = df_tfidf.shape[0]

#Term Frequency
#(Number of times term w appears in a document) / (Total number of 
#terms in the document)

total_words_doc = df_tfidf.astype(bool).sum(axis=1)
tf = df_tfidf.values/total_words_doc[:,None]

#Inverse document frequency
#log_e(Total number of documents / Number of documents with term w in 
#it)
words_in_doc = df_tfidf.astype(bool).sum(axis=0)
idf = np.log(total_docs/words_in_doc)

tf_idf = tf*idf.values[None,:]

return tf_idf

Upvotes: 0

Views: 1351

Answers (2)

krishna_koundinya
krishna_koundinya

Reputation: 11

Use TfidfTransformer. Using TfidfVectorizer on counts might give us undesired results. Let's change the term frequency matrix you provided to understand the difference.

doc apple banana cat
1 2 5 0
2 0 0 3
3 0 6 4

Using TfidfTransformer

from sklearn.feature_extraction.text import TfidfTransformer
transformer = TfidfTransformer()
tfidf_trans = transformer.fit_transform(df)
tfidf_trans_df = pd.DataFrame(tfidf_trans.toarray(), index = df.index, columns=df.columns)
print(tfidf_trans_df)

Output:

doc apple banana cat
1 0.465494 0.885051 0.0
2 0.0 0.0 1.0
3 0.0 0.832050 0.5547

Using TfidfVectorizer

from sklearn.feature_extraction.text import TfidfVectorizer
vectorizer= TfidfVectorizer()
tfidf_vect = vectorizer.fit_transform(df)
tfidf_vect_df = pd.DataFrame(tfidf_vect.toarray(), index = df.index, columns=df.columns)
print(tfidf_vect_df)

Output:

doc apple banana cat
1 1.0 0.0 0.0
2 0.0 1.0 0.0
3 0.0 0.0 1.0

Please refer https://scikit-learn.org/stable/modules/feature_extraction.html#tfidf-term-weighting for a detailed explanation.

Upvotes: 1

ExplodingGayFish
ExplodingGayFish

Reputation: 2897

Suppose you have a Count Vectorizer as a pandas.DataFrame like this:

import pandas as pd
data = [[1,2,0,0],[2,0,0,2],[3,0,2,0]]
df = pd.DataFrame(data,columns=['doc','apple','banana','cat'])
df

Output:

doc apple   banana  cat
0   1   2   0   0
1   2   0   0   2
2   3   0   2   0

Then you can use sklearn.feature_extraction.text.TfidfVectorizer to get the tf-idf vector like this:

from sklearn.feature_extraction.text import TfidfVectorizer
v = TfidfVectorizer()
x = v.fit_transform(df)
df1 = pd.DataFrame(x.toarray(), columns=v.get_feature_names())
print(df1)

Output:

apple  banana  cat  doc
0    0.0     0.0  0.0  1.0
1    1.0     0.0  0.0  0.0
2    0.0     1.0  0.0  0.0
3    0.0     0.0  1.0  0.0

Upvotes: 0

Related Questions