def run(args):
if args.build:
model = build(args)
if args.best:
args.rocchio = True
try:
print('loading model from %s' % args.load_model)
model = VSM(model_path=args.load_model)
except:
print('failed to load model, build from raw.')
model = build(args)
query_list, query_id = process_query(args.query_file)
output_file = open(args.ranked_list, 'w+')
print('query_id,retrieved_docs', file=output_file)
for i, query in enumerate(query_list):
doc_id, doc_score = model.get_ranking(query, args.rocchio)
print('%s,%s' % (query_id[i], ' '.join(doc_id[:100])),
file=output_file)
def run():
import pandas as pd
model = VSM(model_path='./model.pl')
query_list, query_id = process_query('../queries/query-train.xml')
answer = pd.read_csv('../queries/ans_train.csv')[['retrieved_docs']].values
answer = [a[0].strip().split(' ') for a in answer]
for p in product_dict(params):
model.k1 = p['k1']
model.b = p['b']
model.k3 = p['k3']
model.tf_type = p['tf']
model.idf_type = p['idf']
model.doc_len_norm = p['norm']
model._compute_idf()
score = []
for i, query in enumerate(query_list):
doc_id, doc_score = model.get_ranking(query, p['rocchio'], p['n'], p['k'])
score.append(MAP(doc_id[:100], answer[i]))
score = sum(score) / len(score)
model_str = 'k1=%.2f,b=%.2f,k3=%d,idf_type=%s,rocchio=%s,n=%d,k=%d,score=%.5f' % (p['k1'], p['b'], p['k3'], p['idf'], p['rocchio'], p['n'], p['k'], score)
print(model_str)
