Closed lihuiliullh closed 2 years ago
May I know why you don't kge to train MetaQA?
Fbwq = freebasewebquestions. Other names might have been better
We used that code to train embeddings for MetaQA. Since that code isn't scalable, we used libkge for fbwq
In hindsight, it would have been better to use libkge for all datasets
If I remember correctly, in one of the closed issues a config file has been shared. If you can look for it there it would be great
Yes, I checked that issue. https://github.com/malllabiisc/EmbedKGQA/issues/97
But I don't know whether it could achieve the same performance as your model. Because there are many parameters in the file, e.g., learning_rate, max_epochs, batch_size, drop out rate. It would be great if you could share your original file.
Let me look for it and I'll get back to you
On Sat, Nov 20, 2021, 10:10 AM lihuiliullh @.***> wrote:
Yes, I checked that issue. #97 https://github.com/malllabiisc/EmbedKGQA/issues/97
But I don't know whether it could achieve the same performance as your model. Because there are many parameters in the file, e.g., learning_rate, max_epochs, batch_size, drop out rate. It would be great if you could share your original file.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/malllabiisc/EmbedKGQA/issues/108#issuecomment-974592890, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAO6ADYQJSWA7QPIUTNBYU3UM4RDTANCNFSM5INPYVLQ . 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.
Thanks. Also, can you share the command line you use to pre-train MetaQA?
This should be the correct config
complex:
entity_embedder:
dropout: 0.44299429655075073
regularize_weight: 7.830760727899156e-12
relation_embedder:
dropout: -0.4746062345802784
regularize_weight: 1.182876478423781e-10
dataset:
name: fbwq_half
eval:
batch_size: 200
chunk_size: 25000
num_workers: 2
import:
- complex
lookup_embedder:
dim: 400
initialize_args:
normal_:
mean: 0.0
std: 5.8970567449527816e-05
regularize_args:
p: 1
weighted: true
sparse: true
model: complex
negative_sampling:
implementation: batch
num_samples:
o: 7851
s: 2176
shared: true
with_replacement: false
train:
auto_correct: true
batch_size: 1024
loss_arg: .nan
lr_scheduler: ReduceLROnPlateau
lr_scheduler_args:
factor: 0.95
mode: max
patience: 1
threshold: 0.0001
max_epochs: 200
num_workers: 8
optimizer_args:
lr: 0.6560544891789137
type: negative_sampling
valid:
early_stopping:
min_threshold:
epochs: 10
metric_value: 0.1
patience: 10
would you mind share the configure file for training fbwq_half and fbwq_full? Also, why do you name it fbwq_half instead of WebQuestionsSP_half?
Another thing is May I know how you pre-train MetaQA data? Do you use the code in directory "train_embeddings" to learn the embedding? If so, can you share the command of running the main.py with me?