There is no full_Dataset.yaml file in datasets/

[hetolin]

Hi, Thank you for sharing your code.It seems useful. I am using your code to retrain the network.But I found that there is no full_Dataset.yaml file in datasets/.
But in the code, it used the full_Dataset.yaml file:

def load_dataset(batch_size, num_cuda_workers, mode, debug=False, seed=100): datafile = "" if debug: datafile = "datasets/debug_Dataset.yaml" elif mode == "train": datafile = "datasets/full_Dataset.yaml" elif mode == "test": datafile = "datasets/full_test_Dataset.yaml"

There only exists objects.yaml and default_train_test_spilt.yaml in the ../datasets/
Could you provide the yaml files? Thank you very much.

[fbottarel]

The same goes for file full_test_Dataset.yaml, needed to test the paper results @jsll

[hetolin]

@fbottarel the code have train_models_train_views and train_models_holdout_views and holdout_models_holdout_views,so maybe we can consider the train_models_holdout_views and holdout_models_holdout_views as test set while train_models_train_views as the train set
?

[jsll]

First of all thanks for reporting this issues and sorry that I forgot to include them. The reason they weren't in the repo was due to the files being too big. I have now added them to the download_data.sh script. So if you just rerun bash download_data.sh it will first download the dataset splits and later on the actual datasets used. If you have already downloaded the datasets there is no need for you to download them again and just abort the script at that time.

As @hetolin mentioned, the train_models_train_views are the views the model is trained on while the train_models_holdout_views and holdout_models_holdout_views are not used for training. In the paper I report the test score on all of these.

[mingmou123]

hello, the download_data.sh consists of datasets.yaml, debug_Dataset.yaml, full_Dataset.yaml and train_test_dataset.yaml. There is not full_test_Dataset.yaml. Does the train_test_dataset.yaml means full_test_Dataset.yaml?