What parameters do you use to achieve slightly better performance than the number reported in the paper?

[xuetf]

Great Approach. Besides, I am wondering what parameters you use to achieve slightly better performance than the number reported in the paper. I change the learning rate to 0.001 and it achieve 29% recall in the citeulike dataset, which is lower than 33% recall reported in the paper. The parameters is as follows. Hope for your help soon.

model = CML(n_users,
n_items,
features=dense_features,
embed_dim=100,
margin=2.0,
clip_norm=1.1,
master_learning_rate=0.001,
hidden_layer_dim=512,
dropout_rate=0.3,
feature_projection_scaling_factor=1,
feature_l2_reg=0.1,
use_rank_weight=True,
use_cov_loss=True,
cov_loss_weight=1
)

[changun]

Hi xuetf, I realized that there is a bug in the dataset pre-processing step for CiteULike (see commit that fixed the bug: c127de4 ) Unfortunately, with the bug fixed, while I still see improvement over other approaches, I were not able to achieved recall beyond 33%.

…

On Mon, Dec 3, 2018 at 6:46 PM xuetf ***@***.***> wrote: Great Approach. Besides, I am wondering what parameters you use to achieve slightly better performance than the number reported in the paper. I change the learning rate to 0.00 and it achieve 29% recall in the citeulike dataset, which is lower than 33% recall reported in the paper. The parameters is as follows. Hope for your help soon. model = CML(n_users, n_items, features=dense_features, embed_dim=200, margin=2.0, clip_norm=1.1, master_learning_rate=0.001, hidden_layer_dim=512, dropout_rate=0.3, feature_projection_scaling_factor=1, feature_l2_reg=0.1, use_rank_weight=True, use_cov_loss=True, cov_loss_weight=1 ) — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#15>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA1Ff5qRJNGyHgMwDZAVho_dxla-8rIEks5u1eIcgaJpZM4Y_z8w> .

-- Cheng-Kang (Andy) Hsieh UCLA Computer Science Ph.D. Student M: (310) 990-4297

[xuetf]

Hi xuetf, I realized that there is a bug in the dataset pre-processing step for CiteULike (see commit that fixed the bug: c127de4 ) Unfortunately, with the bug fixed, while I still see improvement over other approaches, I were not able to achieved recall beyond 33%.
…
On Mon, Dec 3, 2018 at 6:46 PM xuetf @.***> wrote: Great Approach. Besides, I am wondering what parameters you use to achieve slightly better performance than the number reported in the paper. I change the learning rate to 0.00 and it achieve 29% recall in the citeulike dataset, which is lower than 33% recall reported in the paper. The parameters is as follows. Hope for your help soon. model = CML(n_users, n_items, features=dense_features, embed_dim=200, margin=2.0, clip_norm=1.1, master_learning_rate=0.001, hidden_layer_dim=512, dropout_rate=0.3, feature_projection_scaling_factor=1, feature_l2_reg=0.1, use_rank_weight=True, use_cov_loss=True, cov_loss_weight=1 ) — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#15>, or mute the thread https://github.com/notifications/unsubscribe-auth/AA1Ff5qRJNGyHgMwDZAVho_dxla-8rIEks5u1eIcgaJpZM4Y_z8w .
-- Cheng-Kang (Andy) Hsieh UCLA Computer Science Ph.D. Student M: (310) 990-4297

Thank you for your answering. I've found that the feature pro-processing step has the same problem as what you said for the ratings pre-processing. The first element of tag-item.dat is the number of items that have the tag. After I fix this bug, I get the expected result as what is reported in your paper, i.e., 33%. (Unfortunately, It also relies on the initialization while I haven't set the seed in advance)

Thank you very much!