Weird issue about the rerun

[Chuan1997]

I ran the command on latest small movielens dataset twice _python peagcn_solver_bpr.py --dataset=Movielens --dataset_name=latest-small --num_core=10 --num_feat_core=10 --sampling_strategy=unseen --entity_aware=false --dropout=0 --emb_dim=64 --repr_dim=16 --hidden_size=64 --meta_path_steps=2,2,2,2,2,2,2,2,2 --entity_aware_coff=0.1 --init_eval=true --gpu_idx=0 --runs=5 --epochs=30 --batch_size=1024 --save_every_epoch=5 --metapathtest=false the first time is normal, but the second time I got this: Dataset loaded! Overall HR@5: 0.3803, HR@10: 0.5362, HR@15: 0.6329, HR@20: 0.7010, NDCG@5: 0.2503, NDCG@10: 0.2999, NDCG@15: 0.3246, NDCG@20: 0.3399, AUC: 0.8242, train loss: 259.2449, eval loss: 43.4085 It actually didn't running.

[jaiabhayk]

I ran the command on latest small movielens dataset twice _python peagcn_solver_bpr.py --dataset=Movielens --dataset_name=latest-small --num_core=10 --num_feat_core=10 --sampling_strategy=unseen --entity_aware=false --dropout=0 --emb_dim=64 --repr_dim=16 --hidden_size=64 --meta_path_steps=2,2,2,2,2,2,2,2,2 --entity_aware_coff=0.1 --init_eval=true --gpu_idx=0 --runs=5 --epochs=30 --batch_size=1024 --save_every_epoch=5 --metapathtest=false the first time is normal, but the second time I got this: Dataset loaded! Overall HR@5: 0.3803, HR@10: 0.5362, HR@15: 0.6329, HR@20: 0.7010, NDCG@5: 0.2503, NDCG@10: 0.2999, NDCG@15: 0.3246, NDCG@20: 0.3399, AUC: 0.8242, train loss: 259.2449, eval loss: 43.4085 It actually didn't running.

How long does it take to run for full or small dataset ?

[ecom-research]

I think first time when you run the code, it loads the dataset in a graph manner and saves it on the disk. When you run it the second time it loads the dataset from the disk and starts training. The training for ml small is quite fast.

For bigger data sets it takes relatively more Time.

[ecom-research]

You can turn off the saving/loading of the dataset/graph model. There is option in the code.

[Chuan1997]

Thanks for the guidance. There is another question, when I tried to run the baseline model like kgat(after I run the PEAGNN), the downloading process repeated again. _No loggers found at 'D:\Study\reproduction\PEAGNN\experiments\checkpoint\loggers\Movielenslatest-small\KGAT\BPR{'model_ty\globallogger.pkl' Downloading http://files.grouplens.org/datasets/movielens/ml-latest-small.zip Extracting D:\Study\reproduction\PEAGNN\experiments\checkpoint\data\Movielenslatest-small\raw\ml-latest-small.zip Processing... Data frame not found in D:\Study\reproduction\PEAGNN\experiments\checkpoint\data\Movielenslatest-small\processed! Read from raw data and preprocessing from D:\Study\reproduction\PEAGNN\experiments\checkpoint\data\Movielenslatest-small\raw\ml-latest-small! Raw files not found! Reading directories and actors from api! Get item resources: 7%|█████▌ | 702/9742 [08:44<1:29:49, 1.68it/s] The process is extremely slow, how can I avoid it and speed up the training process?