algue-rythme / ocsdf Goto Github PK
View Code? Open in Web Editor NEWSemi-supervised learning of 1-Lipschitz Signed Distance Functions
License: MIT License
Semi-supervised learning of 1-Lipschitz Signed Distance Functions
License: MIT License
Hi! Thank you very much for your excellent work and generous code sharing.
I am running your code and trying to reproduce the results of your paper.
In the run_mnist.ipynb notebook, when I run the code to train the model, the [test/ood] AUROC remains around 50. Do you have any suggestions on this issue?
I didn't modify most of the code and just commented out the line
plot_gan(epoch, model, p_batch, Q0[:16], gen, maxiter=config.maxiter, **train_kwargs)
Here are some of the results:
Epoch=24 LR=0.0006578
100%|██████████| 47/47 [06:48<00:00, 8.69s/it, FP=0.00%, GN_Qt=0.603, P=3.756, Q0=-8.280, Qt=-1.314, R=1.00%, lipschitz_ratio=1.642, loss=-0.018]
24/24 [==============================] - 6s 265ms/step
4/4 [==============================] - 1s 234ms/step
36/36 [==============================] - 10s 268ms/step
Train examples DescribeResult(nobs=6016, minmax=(1.0218256, 5.828541), mean=3.8103046, variance=0.6504586325222913, skewness=-0.4136563609162037, kurtosis=-0.11905740571011902)
Test examples DescribeResult(nobs=892, minmax=(1.4499859, 5.792158), mean=3.813585, variance=0.5596361719531778, skewness=-0.27269367098282, kurtosis=-0.2819219690014747)
OOD examples DescribeResult(nobs=9108, minmax=(-0.023437515, 6.628998), mean=3.880484, variance=1.299644826991872, skewness=0.14530889035233022, kurtosis=-0.48042250906903483)
[train/ood] Avg Dist=-0.070 T=6.629 Acc=60.22%
[train/ood] AUROC =49.567
[test /ood] Avg Dist =-0.06689906120300293 T=6.629 Acc=91.08%
[test /ood] AUROC =49.439
Epoch=25 LR=0.0006435
100%|██████████| 47/47 [06:47<00:00, 8.67s/it, FP=0.03%, GN_Qt=0.585, P=5.180, Q0=-7.722, Qt=-1.162, R=1.00%, lipschitz_ratio=1.597, loss=-0.001]
24/24 [==============================] - 6s 264ms/step
4/4 [==============================] - 1s 233ms/step
36/36 [==============================] - 10s 267ms/step
Train examples DescribeResult(nobs=6016, minmax=(1.6921498, 5.955354), mean=4.2076674, variance=0.4955580529825745, skewness=-0.47829623738328975, kurtosis=-0.019798706490049334)
Test examples DescribeResult(nobs=892, minmax=(2.1150906, 5.875563), mean=4.2115965, variance=0.42422112698774417, skewness=-0.3225641967050908, kurtosis=-0.2273849390289704)
OOD examples DescribeResult(nobs=9108, minmax=(0.9116055, 6.60164), mean=4.287896, variance=0.9838932968199129, skewness=0.055762059084495164, kurtosis=-0.446357031111134)
[train/ood] Avg Dist=-0.080 T=6.602 Acc=60.22%
[train/ood] AUROC =48.594
[test /ood] Avg Dist =-0.07629966735839844 T=6.602 Acc=91.08%
[test /ood] AUROC =48.417
Epoch=26 LR=0.0006292
100%|██████████| 47/47 [06:42<00:00, 8.57s/it, FP=0.00%, GN_Qt=0.590, P=4.211, Q0=-8.329, Qt=-1.318, R=1.00%, lipschitz_ratio=1.603, loss=-0.023]
24/24 [==============================] - 6s 269ms/step
4/4 [==============================] - 1s 234ms/step
36/36 [==============================] - 10s 269ms/step
Train examples DescribeResult(nobs=6016, minmax=(0.820788, 5.470547), mean=3.5400138, variance=0.603637361724676, skewness=-0.44938489407989984, kurtosis=-0.07952032619838567)
Test examples DescribeResult(nobs=892, minmax=(1.249703, 5.3880486), mean=3.5439668, variance=0.5181393888261584, skewness=-0.30554993974965317, kurtosis=-0.27075616964335314)
OOD examples DescribeResult(nobs=9108, minmax=(-0.16365284, 6.181051), mean=3.6074412, variance=1.2093181372395694, skewness=0.10241750543799147, kurtosis=-0.4865591246523189)
[train/ood] Avg Dist=-0.067 T=6.181 Acc=60.22%
[train/ood] AUROC =49.412
[test /ood] Avg Dist =-0.06347441673278809 T=6.181 Acc=91.08%
[test /ood] AUROC =49.282
Thank you again for your sharing, and I look forward to your reply.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 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. 📊📈🎉
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 ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.