def __calc_mae(self, test_db):
""" This method evaluates that how much our prediction's
are accurate.
This distinction is very important. MAE is able to characterize the
accuracy of prediction not accuracy of recommendation[1].
If you want to evaluate accuracy of prediction, you should
use 'pr' rather than 'mae'. 'pr' provides you Precision and Recall.
[1]: Symeonidis, P. et al., 2006. Collaborative Filtering: Fallacies
and Insights in Measuring Similarity. In citeulikeorg. Citeseer,
p. 56. Available at:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.61.8340&
rep=rep1&type=pdf.
"""
rating_list = []
for user, itemRating in test_db.iteritems():
# user : integer - user_id
# itemRating : dictionary - { item_id:Rating, ... }
rating = self.rec.predict(user, itemRating.keys()[0])
rating_list.append([rating, itemRating.values()[0]])
rating_array = np.array(rating_list)
print "Number of key error: %s" % self.rec.adapt.numberOfKeyError
#TODO: self.eval_metric?
return mae(rating_array[:,0], rating_array[:,1])
def __calc_mae(self, test_db):
""" This method evaluates that how much our prediction's
are accurate.
This distinction is very important. MAE is able to characterize the
accuracy of prediction not accuracy of recommendation[1].
If you want to evaluate accuracy of prediction, you should
use 'pr' rather than 'mae'. 'pr' provides you Precision and Recall.
[1]: Symeonidis, P. et al., 2006. Collaborative Filtering: Fallacies
and Insights in Measuring Similarity. In citeulikeorg. Citeseer,
p. 56. Available at:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.61.8340&
rep=rep1&type=pdf.
"""
rating_list = []
for user, itemRating in test_db.iteritems():
# user : integer - user_id
# itemRating : dictionary - { item_id:Rating, ... }
rating = self.rec.predict(user, itemRating.keys()[0])
rating_list.append([rating, itemRating.values()[0]])
rating_array = np.array(rating_list)
print "Number of key error: %s" % self.rec.adapt.numberOfKeyError
#TODO: self.eval_metric?
return mae(rating_array[:, 0], rating_array[:, 1])
def evaluate(self):
#shuffle(self.rec.X) # First, dataset should be shuffled.
result = []
n = len(self.testX)
for i, item in enumerate(self.testX):
prediction = self.rec.predict(item)
result.append(prediction)
if (i+1) % 32 == 0:
pass
#print "%s/%s completed..." % (i, n)
return mae(result, self.testy)