classicsong
commented
3 years ago This division problem is fixed in master. We now use floor_divide.
Closed jeanimal closed 3 years ago
classicsong
commented
3 years ago This division problem is fixed in master. We now use floor_divide.
The documentation says any python >= 3.5 is fine. This is false. It must be python 3.5 because the behavior of division changed in python 3.6, and running dglke_predict results in Runtime Error, specifically “RuntimeError: Integer division of tensors using div or / is no longer supported, and in a future release div will perform true division as in Python 3. Use true_divide or floor_divide (// in Python) instead.”
P.S. I can upload a conda environment.yml if that would help.
= Using python 3.6 =
When running predict, division is a runtime error.
(dglke) ➜ dgl-ke DGLBACKEND=pytorch dglke_predict --model_path ckpts/TransE_l2_FB15k_0/ --format 'h_r_t' --data_files head.list rel.list tail.list --exec_mode all --score_func logsigmoid --topK 5 ckpts/TransE_l2_FB15k_0/config.json {'dataset': 'FB15k', 'model': 'TransE_l2', 'emb_size': 400, 'max_train_step': 500, 'batch_size': 1000, 'neg_sample_size': 200, 'lr': 0.25, 'gamma': 19.9, 'double_ent': False, 'double_rel': False, 'neg_adversarial_sampling': True, 'adversarial_temperature': 1.0, 'regularization_coef': 1e-09, 'regularization_norm': 3, 'emap_file': 'entities.dict', 'rmap_file': 'relations.dict'} Traceback (most recent call last): File "/opt/miniconda3/envs/dglke/bin/dglke_predict", line 8, in
sys.exit(main())
File "/opt/miniconda3/envs/dglke/lib/python3.6/site-packages/dglke/infer_score.py", line 216, in main
result = model.topK(head, rel, tail, args.exec_mode, args.topK)
File "/opt/miniconda3/envs/dglke/lib/python3.6/site-packages/dglke/models/infer.py", line 148, in topK
idx = idx / num_tail
RuntimeError: Integer division of tensors using div or / is no longer supported, and in a future release div will perform true division as in Python 3. Use true_divide or floor_divide (// in Python) instead.
= Using python 3.5 =
When running predict, division generates a warning-- which you can ignore-- that the behavior of division is changing soon.
(dglke_py3_5) ➜ dglke DGLBACKEND=pytorch dglke_predict --model_path ckpts/TransE_l2_FB15k_0/ --format 'h_r_t' --data_files head.list rel.list tail.list --exec_mode all --score_func logsigmoid --topK 5 Using backend: pytorch ckpts/TransE_l2_FB15k_0/config.json {'gamma': 19.9, 'batch_size': 1000, 'max_train_step': 500, 'emap_file': 'entities.dict', 'neg_sample_size': 200, 'emb_size': 400, 'neg_adversarial_sampling': True, 'rmap_file': 'relations.dict', 'double_rel': False, 'dataset': 'FB15k', 'adversarial_temperature': 1.0, 'model': 'TransE_l2', 'double_ent': False, 'regularization_coef': 1e-09, 'lr': 0.25, 'regularization_norm': 3} /Users/distiller/project/conda/conda-bld/pytorch_1591914879026/work/aten/src/ATen/native/BinaryOps.cpp:81: UserWarning: Integer division of tensors using div or / is deprecated, and in a future release div will perform true division as in Python 3. Use true_divide or floor_divide (// in Python) instead. Inference Done The result is saved in result.tsv
Issue #, if available:
Description of changes:
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.