-
Notifications
You must be signed in to change notification settings - Fork 86
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Evaluation Benchmark on the trained model #16
Comments
Did you load the exact trained model? |
The cosine similarities of returned snippets are super small (it can be 0.4 or so). There might be something wrong. I will also check the problem. |
@guxd any news? I am very eager to re-create the papers results, they are very impressive 😄 |
We tested the code and everything is just normal. |
@guxd thanks for the response. actually, i used |
you should first rename `use.codevecs.nromalized.h5' to 'use.codevecs.h5', then run norm_code_repr.py. I fixed this by incorporating |
@guxd i have renamed it and ran run norm_code_repr.py. Now, the results from the top of the issue are:
for the query:
Since the similarities are still not what i anticipated - were there any more updates in the last commit? thank you for your patience and assistance. |
@guxd should i re-train the model? if so, should i edit something special in configs.py (in addition to turning reload from 500 to 0)? Are there any updates in the data? |
@guxd i re-trained the model. Here are the results i now got for
thanks for your patience & assistance ! |
Still worse than the original results. It might be OK as a baseline. |
I meet the same problem with you. How do you increase the similarity from 0.1 to 0.3? Did you just retrain the model? Does the pre-training model work poorly? |
@skye95git Perhaps you forgot to set the reload hyperparameter in config.py. |
Hi, i am trying to reproduce the papers results. I re-trained the Keras model for 500 epochs (as written in the paper) to replicate the originally returned code snippets. I followed the readme instructions and still couldn't reach the paper's results. I then tried updating the reload param in configs.py to be 500 and got the following responses for queries from the benchmark:
For the query:
iterate through a hashmap
I got the responses:
Where only one was related to hash map (and no iteration)
for the query:
convert an inputstream to a string
I got the responses:
And similiar quality results which don't fit the result table for the other queries.
The text was updated successfully, but these errors were encountered: