Tensorflow version

Question

Tensorflow version

trishav96 opened this issue 6 years ago · 28 comments

trishav96 commented 6 years ago

Hi,

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

Thanks.

Answer 1 · 2019-04-09T16:10:06.000Z

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

Answer 2 · 2019-04-09T16:13:21.000Z

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?

Answer 3 · 2019-04-09T16:14:45.000Z

This should be fine.

Answer 4 · 2019-04-09T16:16:40.000Z

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

Answer 5 · 2019-04-09T17:20:41.000Z

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?

Answer 6 · 2019-04-09T17:22:50.000Z

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.

Answer 7 · 2019-04-09T17:37:10.000Z

Which version of numpy do I use?

Answer 8 · 2019-04-09T17:38:38.000Z

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

Answer 9 · 2019-04-09T23:48:59.000Z

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

Answer 10 · 2019-04-10T00:53:06.000Z

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

Answer 11 · 2019-04-10T01:06:16.000Z

What about the completed knowledge graph?

Answer 12 · 2019-04-10T01:42:58.000Z

No there is not such output.

…

On Tue, Apr 9, 2019 at 6:06 PM trishav96 ***@***.***> wrote: What about the completed knowledge graph? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQQQ7wnMwX01klgFPn3edA68-nrfCz5ks5vfTkIgaJpZM4cjGj_> .

Answer 13 · 2019-04-10T01:45:02.000Z

But the paper title is knowledge graph completion right?

Answer 14 · 2019-04-10T17:42:03.000Z

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 ***@***.***> wrote: No there is not such output. On Tue, Apr 9, 2019 at 6:06 PM trishav96 ***@***.***> wrote: > What about the completed knowledge graph? > > — > You are receiving this because you commented. > > > Reply to this email directly, view it on GitHub > <#11 (comment)>, 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 <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AS_iWPU3ZClHc--Vznj0HnNeaumQ5SPCks5vfUGigaJpZM4cjGj_> .

Answer 15 · 2019-04-11T23:18:26.000Z

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

Answer 16 · 2019-04-12T00:08:56.000Z

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 ***@***.***> 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 <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQQQ-on0XIJd0SeqeUbWVydvv63nva-ks5vf8LCgaJpZM4cjGj_> .

Answer 17 · 2019-04-12T03:55:29.000Z

Are they tensorflow variables or normal variables?

Answer 18 · 2019-04-12T04:14:48.000Z

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)

Answer 19 · 2019-04-12T04:25:41.000Z

Yes!

…

On Thu, Apr 11, 2019 at 9:14 PM trishav96 ***@***.***> 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 <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQQQ6Tuqaqy7iIrfYO7gaMszeWRGo_Yks5vgAg4gaJpZM4cjGj_> .

Answer 20 · 2019-04-12T04:27:03.000Z

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:*)

Answer 21 · 2019-04-12T04:28:11.000Z

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 ***@***.***> 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 <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQQQyyff1H98ApGE01SNtqg6rHTUDefks5vgAsXgaJpZM4cjGj_> .

Answer 22 · 2019-04-12T04:29:13.000Z

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.

Answer 23 · 2019-04-12T04:31:07.000Z

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 ***@***.***> 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 <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQQQ5ymnBiQPvkCtJall4XOT98vtukDks5vgAuZgaJpZM4cjGj_> .

Answer 24 · 2019-04-12T04:31:46.000Z

Okay, I'll try searching it. Thanks.

Answer 25 · 2019-04-13T09:13:16.000Z

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?

Answer 26 · 2019-04-14T18:53:55.000Z

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

Answer 27 · 2019-04-14T19:01:51.000Z

/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.

Answer 28 · 2019-04-14T19:02:14.000Z

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