You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Here, we provide a high-quality reproducing benchmark for the followers of AFN.
AFN+ results:
Criteo
AUC
Logloss
Paper (official)
0.8074
0.4451
Ours (FuxiCTR)
0.8142
0.4377
Improve
+0.0068
-0.0074
Avazu
AUC
Logloss
Paper (official)
0.7555
0.3718
Ours (FuxiCTR)
0.7641
0.3671
Improve
+0.0086
-0.0047
Note that we use the same settings with the paper, but perform exhaustive hyper-parameter search to achieve better results. Meanwhile, we think regularization and dropout are key to attain such improvements.
Create a data directory and put the downloaded csv files in ../data/Criteo/Criteo_x1.
Both dataset_config.yaml and model_config.yaml files are available in AFN+_criteo_x1_tuner_config_04. Make sure the data paths in dataset_config.yaml are correctly set to what we create in the last step.
Here, we provide a high-quality reproducing benchmark for the followers of AFN.
AFN+ results:
Note that we use the same settings with the paper, but perform exhaustive hyper-parameter search to achieve better results. Meanwhile, we think regularization and dropout are key to attain such improvements.
Reproducing Steps
A hands-on guide to run the AFN+ model on the Criteo_x1 dataset.
Copied from: https://github.com/openbenchmark/BARS/tree/master/ctr_prediction/benchmarks/AFN/AFN%2B_criteo_x1
For more results on AFN, please visit https://github.com/openbenchmark/BARS/tree/master/ctr_prediction/benchmarks/AFN
Index
Environments | Dataset | Code | Results | Logs
Environments
Hardware
Software
Dataset
Dataset ID: Criteo_x1. Please refer to the dataset details to get data ready.
Code
We use FuxiCTR-v1.1.0 for this experiment. See the model code: AFN.
Running steps:
Download FuxiCTR-v1.1.0 and install all the dependencies listed in the environments. Then modify run_expid.py to add the FuxiCTR library to system path
Create a data directory and put the downloaded csv files in
../data/Criteo/Criteo_x1
.Both
dataset_config.yaml
andmodel_config.yaml
files are available in AFN+_criteo_x1_tuner_config_04. Make sure the data paths indataset_config.yaml
are correctly set to what we create in the last step.Run the following script to start.
Results
Total 5 runs:
Logs
The text was updated successfully, but these errors were encountered: