IR-W2V-25.py

from sklearn.metrics.pairwise import cosine_similarity

# Function for calculating average precision for a query
def average_precision(qid,qvector):
  
  # Getting the ground truth and document vectors
  qresult=testing_result.loc[testing_result['qid']==qid,['docid','rel']]
  qcorpus=testing_corpus.loc[testing_corpus['docid'].isin(qresult['docid']),['docid','vector']]
  qresult=pd.merge(qresult,qcorpus,on='docid')
  
  # Ranking documents for the query
  qresult['similarity']=qresult['vector'].apply(lambda x: cosine_similarity(np.array(qvector).reshape(1, -1),np.array(x).reshape(1, -1)).item())
  qresult.sort_values(by='similarity',ascending=False,inplace=True)

  # Taking Top 10 documents for the evaluation
  ranking=qresult.head(10)['rel'].values
  
  # Calculating precision
  precision=[]
  for i in range(1,11):
    if ranking[i-1]:
      precision.append(np.sum(ranking[:i])/i)
  
  # If no relevant document in list then return 0
  if precision==[]:
    return 0

  return np.mean(precision)

# Calculating average precision for all queries in the test set
testing_queries['AP']=testing_queries.apply(lambda x: average_precision(x['qid'],x['vector']),axis=1)

# Finding Mean Average Precision
print('Mean Average Precision=>',testing_queries['AP'].mean())