Open Engineerumair opened 5 years ago
I can't find “embeddings_file” in this project: embeddings_file = "./../embdeepwalk/%s%d.deepwalk"%(graph_name, emb_size) model = word2vec.KeyedVectors.load_word2vec_format(embeddings_file, binary=False)
I can't find “embeddings_file” in this project: embeddings_file = "./../embdeepwalk/%s%d.deepwalk"%(graph_name, emb_size) model = word2vec.KeyedVectors.load_word2vec_format(embeddings_file, binary=False)
You should generate embedding file yourself and give that path to the project.
yes, exactly, you need to apply a node to vector model yourself first, then approximate the distance by FFN
you can try the euclidean distance ||v1-v2|| as an additional experiment
yes, exactly, you need to apply a node to vector model yourself first, then approximate the distance by FFN
Fatemeh, I have applied the same work on weighted graphs but the accuracy is very less, I found you considered the shortest nodes of cost between 2 and 6 only as the dataset were having equal distribution of these weights.
In my case the when I assign weighted between 1 and 10 and when tries to generate train and test set with weighted cost the cost exceeds 10 as there are different nodes added to the path with different weights.
I think the embedding trained between 1 and 10 weights where FFN trained on higher cost , may be this is the problem where it can't predict well.
Do you suggest something for weighted graphs to be trained on?
my method is just for unweighted graphs. For the weighted graphs, you should apply a weighted node to vector model first. Then, approximate the distance using other NN models like Siamese networks. Or even think about building your own NN model that works better!
Yeah, I have used weighted graph and generated training pairs for weighted graph using Dijkstra rather than BFS.
Should I use embedding on graph or training set because some of the training set cost are not in weighted graph, what do you suggest.
On Mon, Oct 18, 2021, 1:17 PM Fatemeh Salehi Rizi @.***> wrote:
my method is just for unweighted graphs. For the weighted graphs, you should apply a weighted node to vector model first. Then, approximate the distance using other NN models like Siamese networks. Or even think about building your own NN model that works better!
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fatemehsrz/Shortest_Distance/issues/1#issuecomment-945490225, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACKA6Y2FKTSIXRQ2KBTGFD3UHPJ27ANCNFSM4IXTFXOQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
there are also embedding techniques particularly for weighted graphs, first use them
I have used node2vec+ embedding which is used for weighted graphs.
On Tue, Oct 19, 2021, 3:27 PM Fatemeh Salehi Rizi @.***> wrote:
there are also embedding techniques particularly for weighted graphs, first use them
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fatemehsrz/Shortest_Distance/issues/1#issuecomment-946588386, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACKA6Y4BVDCQXQGXCCASDCDUHVBXZANCNFSM4IXTFXOQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
Is this essential to learn embeddings first on graph then train neural network on training set, shouldn't we train both embedding and neural network on training set?
On Tue, Oct 19, 2021, 6:33 PM Syed Umair Shah @.***> wrote:
I have used node2vec+ embedding which is used for weighted graphs.
On Tue, Oct 19, 2021, 3:27 PM Fatemeh Salehi Rizi < @.***> wrote:
there are also embedding techniques particularly for weighted graphs, first use them
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fatemehsrz/Shortest_Distance/issues/1#issuecomment-946588386, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACKA6Y4BVDCQXQGXCCASDCDUHVBXZANCNFSM4IXTFXOQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
I also get good accuracy when I use training set of cost between 1 and 5.
On Tue, Oct 19, 2021, 10:12 PM Syed Umair Shah @.***> wrote:
Is this essential to learn embeddings first on graph then train neural network on training set, shouldn't we train both embedding and neural network on training set?
On Tue, Oct 19, 2021, 6:33 PM Syed Umair Shah @.***> wrote:
I have used node2vec+ embedding which is used for weighted graphs.
On Tue, Oct 19, 2021, 3:27 PM Fatemeh Salehi Rizi < @.***> wrote:
there are also embedding techniques particularly for weighted graphs, first use them
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fatemehsrz/Shortest_Distance/issues/1#issuecomment-946588386, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACKA6Y4BVDCQXQGXCCASDCDUHVBXZANCNFSM4IXTFXOQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
Can I have 30 minutes voice meeting over zoom to share my results with you?
On Tue, Oct 19, 2021, 10:16 PM Syed Umair Shah @.***> wrote:
I also get good accuracy when I use training set of cost between 1 and 5.
On Tue, Oct 19, 2021, 10:12 PM Syed Umair Shah < @.***> wrote:
Is this essential to learn embeddings first on graph then train neural network on training set, shouldn't we train both embedding and neural network on training set?
On Tue, Oct 19, 2021, 6:33 PM Syed Umair Shah < @.***> wrote:
I have used node2vec+ embedding which is used for weighted graphs.
On Tue, Oct 19, 2021, 3:27 PM Fatemeh Salehi Rizi < @.***> wrote:
there are also embedding techniques particularly for weighted graphs, first use them
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fatemehsrz/Shortest_Distance/issues/1#issuecomment-946588386, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACKA6Y4BVDCQXQGXCCASDCDUHVBXZANCNFSM4IXTFXOQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
I saw the path for test data set in generate_train_test.py is the following :
f2= open('./data/%s_test.txt'%(graph_name), 'w')
But I found the following in feedforward.py : f2= open('%s_truth.txt'%(graph_name), 'r')