Cannot find the propagated labels after Modified Adsorption?

[pvenkat...@gmail.com]

Hello. I am fairly new to this toolkit. 

I'll get right in: I have a semi-supervised style graph. It has 69000 nodes with more than 4 million equally weighted edges.

It consists of 650 seeds which are classified between 2 labels: Say 0 and 1.

I wish to propagate these labels to all the nodes through modified adsorption. I need to classify all the labels into 0s and 1s.

I followed the procedure given in the simple_config file, and successfully generated the output. No gold labels are being used here.

According to the syntax mentioned in simple_config, I see the scores of the seed labels, and then the estimated labels with their scores, then a boolean to check if the particular node is a test label, and then the MRR.

So, I need to know, depending on the weights of the graph, what is the algorithm classifying a particular node into - 0 or 1? Am I supposed to calculate the outcome myself based on the scores given in the output? Or is it there and I just can't see it? 

I'm sorry if this sounds simple or silly. I just have no one else I can approach. Thanks for your time. Your help is appreciated.

I can give you any more details on query as fast as I can. Thanks.

[Partha Talukdar]

On Thu, Jun 16, 2016 at 6:41 PM, <pvenkat...@gmail.com> wrote:

Am I supposed to calculate the outcome myself based on the scores given in the output?

yes. 

[pvenkat...@gmail.com]

Thank you for your quick reply sir. So, to the unlabeled node, assign the label that has the most influence on it.

By the way, thank you very much for your work on this toolkit sir. This is the only toolkit that implements the algorithms I am regularly in need of, and as far as I know there is no one who did such an extensive research in this field like you did. We are grateful. 

[Partha Talukdar]

On Thu, Jun 16, 2016 at 9:11 PM, <pvenkat...@gmail.com> wrote:

So, to the unlabeled node, assign the label that has the most influence on it.

Yes, that's the usual way to do it.