tag:blogger.com,1999:blog-3211409948956809184.post2242550673642580719..comments2024-03-21T04:14:27.443-07:00Comments on Large Scale Machine Learning and Other Animals: Label propagation in GraphChiDanny Bicksonhttp://www.blogger.com/profile/01517237836051035400noreply@blogger.comBlogger5125tag:blogger.com,1999:blog-3211409948956809184.post-86201851500231076902014-03-25T06:57:19.447-07:002014-03-25T06:57:19.447-07:00Hi Ivan,
You are perfectly right, we output by ro...Hi Ivan, <br />You are perfectly right, we output by rows and not by columns. <br /><br />ThanksDanny Bicksonhttps://www.blogger.com/profile/01517237836051035400noreply@blogger.comtag:blogger.com,1999:blog-3211409948956809184.post-60592486142761998482014-03-25T06:53:50.760-07:002014-03-25T06:53:50.760-07:00Hello, thank you for the implementation of the LP ...Hello, thank you for the implementation of the LP algorithm in Graphchi and for these examples.<br />In the output, columns and rows seems to be inverted. MatrixMarket array format lists the entries by columns. However, the first 4 entries in renren_U.mm correspond to the first ROW of the matrix according to the header "5 4" which defines 5 rows and 4 columns and correspond to the probability distribution over the 4 classes for node 1. Am I wrong? Ivanhttps://www.blogger.com/profile/10958747792938495505noreply@blogger.comtag:blogger.com,1999:blog-3211409948956809184.post-33465618274514058202014-01-12T06:57:13.328-08:002014-01-12T06:57:13.328-08:00Hi Danny, thanks for the tutorial on Label propaga...Hi Danny, thanks for the tutorial on Label propagation. I have a question about high dimensionality. Does reducing the dimensions of a high dimensional Y matrix using say for example PCA effect the working of the algorithm. I have a high dimensional data set and was wondering do I need to reduce the dimensions before running your tool for good results?Riyazhttps://www.blogger.com/profile/14805451207008418628noreply@blogger.comtag:blogger.com,1999:blog-3211409948956809184.post-62519018631929273842014-01-04T21:55:56.014-08:002014-01-04T21:55:56.014-08:00As can be seen in the code: https://github.com/Gra...As can be seen in the code: https://github.com/GraphChi/graphchi-cpp/blob/master/toolkits/graph_analytics/label_propagation.cpp#L151<br />non-seed nodes get a random initial label probability and thus you get results for classes which are not observed. You can change this behavior to uniform probability if you lie.Danny Bicksonhttps://www.blogger.com/profile/01517237836051035400noreply@blogger.comtag:blogger.com,1999:blog-3211409948956809184.post-10381984028576210812013-10-31T05:28:09.523-07:002013-10-31T05:28:09.523-07:00Hi Danny,
If the only labeled instance in the exa...Hi Danny, <br />If the only labeled instance in the example does not have any probability assigned to class 4, how does the output assign weights to class 4, for example, 2.033766122622e-06 on instance 2?Pedro Calaishttps://www.blogger.com/profile/16052031087424796826noreply@blogger.com