Tensorflow version

[trishav96]

Hi,

Which version of tensorflow and python is used in the project?

Thanks.

[bxshi]

Hi I think it is Python3 and an older version of TensorFlow maybe 1.3.

[trishav96]

Okay. And the command can be this? ./ProjE_softmax.py --dim 5 --batch 5 --data ./data/FB15k/ --eval_per 1 --worker 5 --eval_batch 5 --max_iter 5 --generator 5?

[bxshi]

This should be fine.

[bxshi]

You could start with the command I shared in the README, that should work.

[trishav96]

It takes too long, so I thought I'll give smaller values. How many CPU cores are ideally needed? And do I need a GPU?

[bxshi]

You don't need a GPU -- I ran this code on a 48 core machine. If you don't have that many cores, you can reduce the number of generators because each of them is a thread. I would suggest you enlarge batch size set max_iter to 1, and reduce generator to 1.

[trishav96]

Which version of numpy do I use?

[bxshi]

Sorry, I don't recall. Something released before 2017 would be fine.

[trishav96]

What is final output of ProjE_softmax.py supposded to be? Given input is a knowledge graph (FB15k).

[bxshi]

Its just the HITS and MR scores printed on the screen.

[trishav96]

What about the completed knowledge graph?

[bxshi]

No there is not such output.

On Tue, Apr 9, 2019 at 6:06 PM trishav96 notifications@github.com wrote:

What about the completed knowledge graph?

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/bxshi/ProjE/issues/11#issuecomment-481494414, or mute the thread https://github.com/notifications/unsubscribe-auth/ABQQQ7wnMwX01klgFPn3edA68-nrfCz5ks5vfTkIgaJpZM4cjGj_ .

[trishav96]

But the paper title is knowledge graph completion right?

[trishav96]

Hey,

Could you explain in a few lines what exactly you are trying to do in the code?

Thanks, Trisha

On Tue, Apr 9, 2019, 6:42 PM bxshi notifications@github.com wrote:

No there is not such output.

On Tue, Apr 9, 2019 at 6:06 PM trishav96 notifications@github.com wrote:

What about the completed knowledge graph?

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/bxshi/ProjE/issues/11#issuecomment-481494414, or mute the thread < https://github.com/notifications/unsubscribe-auth/ABQQQ7wnMwX01klgFPn3edA68-nrfCz5ks5vfTkIgaJpZM4cjGj_

.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/bxshi/ProjE/issues/11#issuecomment-481500824, or mute the thread https://github.com/notifications/unsubscribe-auth/AS_iWPU3ZClHc--Vznj0HnNeaumQ5SPCks5vfUGigaJpZM4cjGj_ .

[trishav96]

Hi, could you tell me which lines have the variables that store the embeddings of the entities in the Proje_softmax.py file?

[bxshi]

I’m out in the woods and dont have access to a computer... it should be named something like entities relations, etc.

On Thu, Apr 11, 2019 at 4:18 PM trishav96 notifications@github.com wrote:

Hi, could you tell me which lines have the variables that store the embeddings of the entities in the Proje_softmax.py file?

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/bxshi/ProjE/issues/11#issuecomment-482369501, or mute the thread https://github.com/notifications/unsubscribe-auth/ABQQQ-on0XIJd0SeqeUbWVydvv63nva-ks5vf8LCgaJpZM4cjGj_ .

[trishav96]

Are they tensorflow variables or normal variables?

[trishav96]

self.__ent_embedding = tf.get_variable("ent_embedding", [self.__n_entity, embed_dim], initializer=tf.random_uniform_initializer(minval=-bound, maxval=bound, seed=345))

Is it this? (Line 187)

[bxshi]

Yes!

On Thu, Apr 11, 2019 at 9:14 PM trishav96 notifications@github.com wrote:

self.__ent_embedding = tf.get_variable("ent_embedding", [self.__n_entity, embed_dim], initializer=tf.random_uniform_initializer(minval=-bound, maxval=bound, seed=345))

Is it this? (Line 187)

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/bxshi/ProjE/issues/11#issuecomment-482430797, or mute the thread https://github.com/notifications/unsubscribe-auth/ABQQQ6Tuqaqy7iIrfYO7gaMszeWRGo_Yks5vgAg4gaJpZM4cjGj_ .

[trishav96]

How to I read this? When I do print(self.__ent_embedding) it print this --> <tf.Variable 'ent_embedding:0' shape=(14951, 1) dtype=float32_ref> Tensor("ProjE_1/TopKV2_1:1", shape=(?, 14951), dtype=int32, device=/device:CPU:*)

[bxshi]

You will need to session.run it. Print it directly just prints out the tensorflow def.

On Thu, Apr 11, 2019 at 9:27 PM trishav96 notifications@github.com wrote:

How to I read this? When I do print(self.__ent_embedding) it print this --> <tf.Variable 'ent_embedding:0' shape=(14951, 1) dtype=float32_ref> Tensor("ProjE_1/TopKV2_1:1", shape=(?, 14951), dtype=int32, device=/device:CPU:*)

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/bxshi/ProjE/issues/11#issuecomment-482432613, or mute the thread https://github.com/notifications/unsubscribe-auth/ABQQQyyff1H98ApGE01SNtqg6rHTUDefks5vgAsXgaJpZM4cjGj_ .

[trishav96]

I've tried doing it. It didn't work. Do you have a link I can follow to do it? Please provide it? It'll be really helpful.

[bxshi]

It’s basic TensorFlow operations. It should return a numpy object. I don’t have any tutorial on hand unfortunately

On Thu, Apr 11, 2019 at 9:29 PM trishav96 notifications@github.com wrote:

I've tried doing it. It didn't work. Do you have a link I can follow to do it? Please provide it? It'll be really helpful.

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/bxshi/ProjE/issues/11#issuecomment-482432932, or mute the thread https://github.com/notifications/unsubscribe-auth/ABQQQ5ymnBiQPvkCtJall4XOT98vtukDks5vgAuZgaJpZM4cjGj_ .

[trishav96]

Okay, I'll try searching it. Thanks.

[trishav96]

Hi,

So I ran the code for a smaller dataset and on 8 cores with 20GB RAM. 1 generator and 2 workers. But the code doesn't end at all. It shows evaluation metrics for last iteration and is just hung there. Am I missing anything?

[bxshi]

It may just take a very long time to finish I guess. I never experienced such a thing before.

[trishav96]

/ProjE_softmax.py --dim 200 --batch 200 --data ./data/FB15k/ --eval_per 10 --worker 3 --eval_batch 500 --max_iter 10 --generator 1

This is the query I ran, I modified eval_per to 10 and max_iter to 10.

[trishav96]

How long would it take? The dataset has ~2.5k triples.