search

ylqfp / recsyscode

Automatically exported from code.google.com/p/recsyscode
0 stars 0 forks source link

fix the bug with svd++ model #2

Open GoogleCodeExporter opened 9 years ago

GoogleCodeExporter commented 9 years ago 
There are bugs with svd++ model. 
When I used the netflix dataset to test the model, it generated nan error. I am 
fixing the bug now.

Original issue reported on code.google.com by honglianglv on 19 May 2011 at 5:21

GoogleCodeExporter commented 9 years ago 
I tried the svd++ but while executing the dot function occurs to have "-inf" as 
a value for qlocal[i] and nan for p[i]. (in the common.cpp file row 26)

Original comment by baggio.a...@gmail.com on 14 Oct 2012 at 10:37

GoogleCodeExporter commented 9 years ago 
修改后代码仍然存在nan的问题，能否提示一下哪里处理的不��
�会发生此问题？

Original comment by kkli...@gmail.com on 18 Apr 2013 at 2:25

GoogleCodeExporter commented 9 years ago 
In svd++ model, you used wrong formulation (i.e., r_{ui} = \mu + \b_u + \b_i + 
p_u \cdot p_i) to predict rating. You should include implicit feedback in this 
formulation as well.

Original comment by jingwei....@gmail.com on 15 May 2015 at 10:12