47
votes

I have a Sqlite database that contains following type of schema:

termcount(doc_num, term , count)

This table contains terms with their respective counts in the document. like

(doc1 , term1 ,12)
(doc1, term 22, 2)
.
.
(docn,term1 , 10)

This matrix can be considered as sparse matrix as each documents contains very few terms that will have a non-zero value.

How would I create a dense matrix from this sparse matrix using numpy as I have to calculate the similarity among documents using cosine similarity.

This dense matrix will look like a table that have docid as the first column and all the terms will be listed as the first row.and remaining cells will contain counts.

2

2 Answers

92
votes
 from scipy.sparse import csr_matrix
 A = csr_matrix([[1,0,2],[0,3,0]])
 >>>A
 <2x3 sparse matrix of type '<type 'numpy.int64'>'
    with 3 stored elements in Compressed Sparse Row format>
 >>> A.todense()
   matrix([[1, 0, 2],
           [0, 3, 0]])
 >>> A.toarray()
      array([[1, 0, 2],
            [0, 3, 0]])

this is an example of how to convert a sparse matrix to a dense matrix taken from scipy

10
votes

I solved this problem using Pandas. Because we want to keep the document ids and term ids.

from pandas import DataFrame 

# A sparse matrix in dictionary form (can be a SQLite database). Tuples contains doc_id        and term_id. 
doc_term_dict={('d1','t1'):12, ('d2','t3'):10, ('d3','t2'):5}

#extract all unique documents and terms ids and intialize a empty dataframe.
rows = set([d for (d,t) in doc_term_dict.keys()])  
cols = set([t for (d,t) in doc_term_dict.keys()])
df = DataFrame(index = rows, columns = cols )
df = df.fillna(0)

#assign all nonzero values in dataframe
for key, value in doc_term_dict.items():
    df[key[1]][key[0]] = value   

print df

Output:

    t2  t3  t1
d2  0  10   0
d3  5   0   0
d1  0   0  12