Skip to content
New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Missing data file #2

Open
noamtzuk opened this issue May 3, 2021 · 1 comment
Open

Missing data file #2

noamtzuk opened this issue May 3, 2021 · 1 comment

Comments

@noamtzuk
Copy link

noamtzuk commented May 3, 2021

Hi,
It seems like you tested pfoodreq_similar_recipes with test_qas_150820 file, but the only file I see in the attached data is test_qas_090820 and there are no references for it in the code. test_qas_150820 is missing.

Can you please help?
Thanks.

@diana-nurbakova
Copy link

Hello,

I don't know if it still makes sense for you @noamtzuk but you can get this file here. This link was provided in the readme.

You will also need these files (also available in the same folder): dish_info_map.json, new_reciptor_507834_tuned_emb.pkl and name_foodkgid_recipe1Mid_foodcomurl_mapping.tab. You will have to modify the path to them in pfoodreq_similar_recipes.yml.

What I'm not sure is what files shall we use as:

train_data: null
valid_data: null
test_data: null

Shall we take the ones from BAMnet/data/kbqa? You might also need to change the value of vocab_size though.
I tried that but the results I've got had a much worse quality:

max KBQA scores found in all examples: 60.579612731933594
min KBQA scores found in all examples: -16.472976684570312
min similarity distance found in all examples: 1.7881393432617188e-07

Question type: constraint
 Recall: 0.3583333345401701
 Precision: 0.4612319408677377
 F1: 0.29678629349872493
MAP: 0.29001073860789944
MAP@1: 0.5741261550823624
MAP@3: 0.3663452524440873
MAR: 0.27892970331425787
MAR@1: 0.21161585834024468
MAR@3: 0.25925735686512275

Question type: overall
 Recall: 0.3583333345401701
 Precision: 0.4612319408677377
 F1: 0.29678629349872493

So I'm not sure it was a good way to deal with that. If you have any suggestions, I'm very willing to hear them.

Thanks.

# for free to join this conversation on GitHub. Already have an account? # to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants