search

SpaceLearner / SessionRec-pytorch

Session-based Recommendation
MIT License
60 stars 12 forks source link

I cannot reproduce the performance of the model in the paper #3

Closed Nishikata97 closed 2 years ago

Nishikata97 commented 2 years ago

First I used environment.yaml to configure the environment and start.sh to test on the diginetica dataset, here are the results for the last four epochs: 20%|██ | 6/30 [14:52<59:37, 149.08s/it] Epoch 5: MRR = 19.008%, Hit = 55.406% Batch 8500: Loss = 4.0826, Time Elapsed = 21.61s Batch 8600: Loss = 4.0721, Time Elapsed = 9.60s Batch 8700: Loss = 4.0687, Time Elapsed = 9.53s Batch 8800: Loss = 4.0753, Time Elapsed = 9.51s Batch 8900: Loss = 4.0801, Time Elapsed = 10.03s Batch 9000: Loss = 4.0732, Time Elapsed = 9.77s Batch 9100: Loss = 4.0731, Time Elapsed = 9.63s Batch 9200: Loss = 4.0772, Time Elapsed = 9.47s Batch 9300: Loss = 4.0917, Time Elapsed = 10.03s Batch 9400: Loss = 4.0949, Time Elapsed = 9.88s Batch 9500: Loss = 4.0945, Time Elapsed = 9.58s Batch 9600: Loss = 4.0916, Time Elapsed = 9.70s Batch 9700: Loss = 4.0954, Time Elapsed = 9.76s Batch 9800: Loss = 4.1006, Time Elapsed = 9.96s 23%|██▎ | 7/30 [17:21<57:01, 148.75s/it]Epoch 6: MRR = 19.013%, Hit = 55.383% Batch 9900: Loss = 4.0622, Time Elapsed = 21.49s Batch 10000: Loss = 4.0686, Time Elapsed = 9.53s Batch 10100: Loss = 4.0630, Time Elapsed = 9.81s Batch 10200: Loss = 4.0753, Time Elapsed = 9.31s Batch 10300: Loss = 4.0691, Time Elapsed = 9.69s Batch 10400: Loss = 4.0818, Time Elapsed = 9.66s Batch 10500: Loss = 4.0660, Time Elapsed = 9.58s Batch 10600: Loss = 4.0944, Time Elapsed = 9.87s Batch 10700: Loss = 4.0769, Time Elapsed = 9.77s Batch 10800: Loss = 4.0792, Time Elapsed = 9.62s Batch 10900: Loss = 4.0737, Time Elapsed = 9.85s Batch 11000: Loss = 4.0862, Time Elapsed = 9.36s Batch 11100: Loss = 4.0784, Time Elapsed = 9.50s Batch 11200: Loss = 4.0959, Time Elapsed = 10.21s Epoch 7: MRR = 19.006%, Hit = 55.379% 27%|██▋ | 8/30 [19:48<54:23, 148.35s/it]Batch 11300: Loss = 4.0781, Time Elapsed = 21.73s Batch 11400: Loss = 4.0772, Time Elapsed = 9.75s Batch 11500: Loss = 4.0463, Time Elapsed = 9.50s Batch 11600: Loss = 4.0733, Time Elapsed = 9.85s Batch 11700: Loss = 4.0664, Time Elapsed = 9.95s Batch 11800: Loss = 4.0548, Time Elapsed = 9.84s Batch 11900: Loss = 4.0820, Time Elapsed = 9.80s Batch 12000: Loss = 4.0775, Time Elapsed = 9.51s Batch 12100: Loss = 4.0711, Time Elapsed = 9.55s Batch 12200: Loss = 4.0749, Time Elapsed = 9.63s Batch 12300: Loss = 4.0626, Time Elapsed = 9.86s Batch 12400: Loss = 4.0845, Time Elapsed = 9.82s Batch 12500: Loss = 4.0734, Time Elapsed = 9.45s Batch 12600: Loss = 4.0856, Time Elapsed = 9.66s 27%|██▋ | 8/30 [22:16<1:01:14, 167.02s/it] Epoch 8: MRR = 19.006%, Hit = 55.379% MRR@20 HR@20 19.015% 55.406%

Process finished with exit code 0

SpaceLearner commented 2 years ago

You can try first clone the repo again and add "--extra" and "--fusion" on the command line of start.sh and increase the order to get better performance.