pg2455 / hybrid-learn2branch Goto Github PK
View Code? Open in Web Editor NEWHybrid Models for Learning to Branch (NeurIPS 2020)
Home Page: https://arxiv.org/abs/2006.15212
License: MIT License
Hybrid Models for Learning to Branch (NeurIPS 2020)
Home Page: https://arxiv.org/abs/2006.15212
License: MIT License
Like the title, the accuracy is pretty low for combinatorial auction, which is the same as ecole/learn2branch. Is this situation normal?
[2021-10-21 20:12:46.586854] EPOCH 253...
[2021-10-21 20:13:00.615919] TRAIN LOSS: 2.607 acc@1: 0.474 acc@3: 0.708 acc@5: 0.816 acc@10: 0.934
[2021-10-21 20:13:16.086814] VALID LOSS: 2.611 acc@1: 0.474 acc@3: 0.703 acc@5: 0.812 acc@10: 0.931
[2021-10-21 20:13:16.087240] EPOCH 254...
[2021-10-21 20:13:29.891938] TRAIN LOSS: 2.604 acc@1: 0.474 acc@3: 0.711 acc@5: 0.818 acc@10: 0.933
[2021-10-21 20:13:45.182254] VALID LOSS: 2.611 acc@1: 0.474 acc@3: 0.703 acc@5: 0.812 acc@10: 0.930
[2021-10-21 20:13:45.182633] EPOCH 255...
[2021-10-21 20:13:58.630930] TRAIN LOSS: 2.593 acc@1: 0.473 acc@3: 0.703 acc@5: 0.815 acc@10: 0.930
[2021-10-21 20:14:13.781607] VALID LOSS: 2.611 acc@1: 0.473 acc@3: 0.703 acc@5: 0.812 acc@10: 0.931
[2021-10-21 20:14:13.782002] EPOCH 256...
[2021-10-21 20:14:27.094316] TRAIN LOSS: 2.591 acc@1: 0.478 acc@3: 0.706 acc@5: 0.816 acc@10: 0.934
[2021-10-21 20:14:42.387882] VALID LOSS: 2.611 acc@1: 0.473 acc@3: 0.703 acc@5: 0.812 acc@10: 0.930
[2021-10-21 20:14:42.388286] EPOCH 257...
[2021-10-21 20:14:55.826369] TRAIN LOSS: 2.579 acc@1: 0.480 acc@3: 0.708 acc@5: 0.817 acc@10: 0.935
[2021-10-21 20:15:10.879072] VALID LOSS: 2.611 acc@1: 0.474 acc@3: 0.703 acc@5: 0.812 acc@10: 0.930
[2021-10-21 20:15:10.879466] EPOCH 258...
[2021-10-21 20:15:24.284918] TRAIN LOSS: 2.575 acc@1: 0.473 acc@3: 0.716 acc@5: 0.820 acc@10: 0.939
[2021-10-21 20:15:39.623349] VALID LOSS: 2.612 acc@1: 0.475 acc@3: 0.703 acc@5: 0.812 acc@10: 0.930
[2021-10-21 20:15:39.623798] EPOCH 259...
[2021-10-21 20:15:53.617875] TRAIN LOSS: 2.597 acc@1: 0.475 acc@3: 0.707 acc@5: 0.812 acc@10: 0.935
[2021-10-21 20:16:09.005999] VALID LOSS: 2.611 acc@1: 0.474 acc@3: 0.703 acc@5: 0.812 acc@10: 0.930
[2021-10-21 20:16:09.006399] EPOCH 260...
[2021-10-21 20:16:22.500397] TRAIN LOSS: 2.576 acc@1: 0.461 acc@3: 0.705 acc@5: 0.810 acc@10: 0.935
[2021-10-21 20:16:38.666407] VALID LOSS: 2.612 acc@1: 0.475 acc@3: 0.704 acc@5: 0.812 acc@10: 0.930
[2021-10-21 20:16:38.666809] EPOCH 261...
[2021-10-21 20:16:52.327668] TRAIN LOSS: 2.598 acc@1: 0.469 acc@3: 0.703 acc@5: 0.818 acc@10: 0.932
[2021-10-21 20:17:06.524168] VALID LOSS: 2.611 acc@1: 0.475 acc@3: 0.704 acc@5: 0.812 acc@10: 0.930
[2021-10-21 20:17:06.524482] EPOCH 262...
[2021-10-21 20:17:19.644947] TRAIN LOSS: 2.595 acc@1: 0.485 acc@3: 0.711 acc@5: 0.818 acc@10: 0.934
[2021-10-21 20:17:35.656261] VALID LOSS: 2.611 acc@1: 0.474 acc@3: 0.703 acc@5: 0.812 acc@10: 0.930
[2021-10-21 20:17:35.656523] 30 epochs without improvement, early stopping
[2021-10-21 20:17:50.041247] BEST VALID LOSS: 2.611 acc@1: 0.474 acc@3: 0.704 acc@5: 0.812 acc@10: 0.931
I tried to generate some samples, get this error at "self.map = sorted([x.getCol().getIndex() for x in self.model.getVars(transformed=True)])".
Hi,
It's a good job and so kind of you to share the code on the Github.
I have run and trained the model following the instruction, and choose the best performing model to evaluate on Setcover problem.
However, I found that the performance is so poor that it can hardly solve one of the big instance(all exceed the limit of time).
I wonder to know whether the situation is normal, or maybe something is wrong with the progress of my training?
The result can be attached as the following
learn2branch_CPU_20221023-214346.csv
hybrid_CPU_20221025-232556.csv
Thank you and have a good day.
Hi,
I really liked your paper and thank you for publicly releasing your codebase.
I wondered if there is any way to have access to the final trained weights used to produce the tables in the paper else than redoing the training from scratch ? This would help me very much help in accurately comparing my technique to this one.
If ever the trained weights are available somewhere that I missed, please excuse me for missing it and just point me towards where to get them.
Thank you and have a good day.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.