+
Skip to content

bjourne/ml-stuff

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Reimplementations of deep learning models.

Usage

$ python mlstuff/scripts.py --help
...
$ python mlstuff/scripts.py --batch-size 128 vgg16 cifar100 train /tmp/logs
...
$ python mlstuff/scripts.py --batch-size 128 vgg16 cifar100 test /path/to/net.pth
...
$ CUDA_VISIBLE_DEVICES=-1 python mlstuff/scripts.py \
	--batch-size 128 vgg16qcfs cifar100 train /tmp/logs

Tests

$ PYTHONPATH=. pytest -vs

Observations

  • With dropout 0.5 VGG16 does not converge on CIFAR100 with learning rate 0.1 and batch size 32.

Results

Abbreviations:

ABBR MEANING
ACC Validation accuracy
AUG Data augmentation strategy
BS Batch size
DO Dropout
LR Learning rate
VER Version code
WD Weight decay
PRG Training in progress
N Number of epochs
DATE DATASET MODEL VER BS AUG WD DO LR SEED N ACC PRG
2025-02-19 CIFAR10 AlexNet std 128/8 aa 0.0 0.0 0.1 1000 976 94.3 n
2025-05-22 CIFAR10 AlexNet std 512/8 aa 0.0005 0.5 0.08 2338 y
2025-05-21 CIFAR10 AlexNet std 1024/8 aa 0.0005 0.5 0.05 5146 433 92.7 n
2025-05-21 CIFAR10 AlexNet std 1024/8 aa 0.0005 0.5 0.01 6286 878 90.4 n
CIFAR10 DenseNet std 128 std 0.0 94.9 n
CIFAR10 DenseNet std 256 std 0.0 94.6 n
CIFAR10 DenseNet std 512 std 0.0 93.8 n
2025-02-11 CIFAR10 DenseNet std 256/8 aa 0.0 0.0 1003 443 96.4 n
2025-05-12 CIFAR10 ResNet18 std 256/8 aa 0.0005 0.0 0.10 3049 981 97.8 n
2025-05-12 CIFAR10 ResNet18 std 256/8 aa 0.0005 0.0 0.20 6619 840 97.8 n
2025-05-12 CIFAR10 ResNet18QCFS std 1024/8 aa 0.0005 0.0 0.20 6271 97.5 n
CIFAR10 VGG16 std ? std 0.0 0.5 93.6 n
CIFAR10 VGG16 lin-512 ? std 0.0 0.5 92.1 n
CIFAR10 VGG16 no-BN ? std 0.0 0.5 92.6 n
2025-05-12 CIFAR10 VGG16 std 1024/8 std 0.0005 0.0 0.20 7473 96.7 n
2025-05-12 CIFAR10 VGG16QCFS std 1024/8 std 0.0005 0.0 0.20 7397 850 97.4 n
2025-05-07 CIFAR100 AlexNet std 256/8 aa 0.0000 0.5 0.10 9850 790 77.6 n
2025-05-07 CIFAR100 AlexNet std 256/8 aa 0.0005 0.5 0.10 9270 925 79.0 n
2025-05-21 CIFAR100 AlexNet std 256/8 aa 0.0005 0.5 0.10 2525 917 78.5 n
2025-05-09 CIFAR100 AlexNetQCFS std 256/8 aa 0.0005 0.5 0.10 3712 489 78.3 n
2025-05-07 CIFAR100 ConvNeXt std 256/8 aa 0.0000 - - 6420 968 71.5 n
2025-02-11 CIFAR100 DenseNet std 256/8 aa 0.0 0.0 1007 415 78.4 n
2025-02-15 CIFAR100 EffNet-B0 std 256/8 aa 0.0 0.0 1010 201 75.5 n
2025-02-15 CIFAR100 EffNet-B0 std 512/8 aa 0.0 0.0 1010 516 77.4 n
2025-02-18 CIFAR100 ResNet18 std 128/8 aa 0.0000 0.0 0.10 1000 710 76.8 n
2025-02-11 CIFAR100 ResNet18 std 256/8 aa 0.0 0.0 1004 731 76.7 n
2025-02-14 CIFAR100 ResNet18 std 256/8 aa 0.0004 0.0 1012 908 77.1 n
2025-02-14 CIFAR100 ResNet18 std 256/8 aa 0.0005 0.0 1010 765 77.7 n
2025-02-14 CIFAR100 ResNet18 std 256/8 aa 0.0005 0.0 1011 923 77.7 n
2025-02-14 CIFAR100 ResNet18 std 256/8 aa 0.0010 0.0 1015 637 75.0 n
2025-05-09 CIFAR100 ResNet18 std 256/8 aa 0.0005 0.0 0.10 4855 1000 80.8 n
2025-02-13 CIFAR100 ResNet18 std 512/8 aa 0.0 0.0 1010 636 76.2 n
2025-02-11 CIFAR100 ResNet18 std 512/8 aa 0.0005 0.0 1003 555 77.3 n
2025-02-11 CIFAR100 ResNet18 std 512/8 aa 0.0010 0.0 1010 619 76.8 n
2025-02-12 CIFAR100 ResNet18 std 1024/8 aa 0.0 0.0 1010 572 75.6 n
2025-02-12 CIFAR100 ResNet18 std 1024/8 aa 0.0010 0.0 1010 771 77.2 n
2025-02-12 CIFAR100 ResNet18 std 1024/8 aa 0.0020 0.0 1010 554 74.2 n
CIFAR100 ResNet50 std 256 aa 0.0 47.8 n
CIFAR100 ResNet20 std 256 aa 0.0 1001 67.2 n
2024-10-01 CIFAR100 ResNet20 std 512 aa 0.0 0.0 1001 69.1 n
2024-10-01 CIFAR100 ResNet20 std 256 aa 0.0 0.0 1001 69.4 n
2025-02-17 CIFAR100 ResNet20 std 128/8 aa 0.0005 0.0 0.10 1000 953 72.8 n
2025-02-18 CIFAR100 ResNet20-PA std 128/8 aa 0.0005 0.0 0.05 1000 962 71.6 n
2025-02-18 CIFAR100 ResNet20-PA std 128/8 aa 0.0005 0.0 0.10 1000 975 71.7 n
2025-02-18 CIFAR100 ResNet20-PA std 128/8 aa 0.0005 0.0 0.10 1001 739 71.0 n
2024-10-06 CIFAR100 ResNet18QCFS std 128 aa 0.0005 0.0 1001 80.3 n
2024-10-06 CIFAR100 ResNet18QCFS std 256 aa 0.0005 0.0 1001 79.8 n
2025-05-09 CIFAR100 ResNet18QCFS std 256/8 aa 0.0005 0.0 0.10 9418 950 82.0 n
CIFAR100 VGG16 std 32 aa 0.0 0.0 1001 74.9 n
CIFAR100 VGG16 std 64 aa 0.0 0.5 1001 69.1 n
CIFAR100 VGG16 std 64 std 0.0 0.5 71.7 n
CIFAR100 VGG16 std 128 aa 0.0 0.5 1001 75.4 n
2025-02-01 CIFAR100 VGG16 std 128 aa 0.0005 0.0 1001 77.4 n
CIFAR100 VGG16 std 256 aa 0.0 0.5 74.7 n
2024-10-02 CIFAR100 VGG16 std 256 aa 0.0005 0.0 1001 77.6 n
2025-05-05 CIFAR100 VGG16 std 256/2 aa 0.0005 0.0 0.10 8330 1000 77.6 n
2025-05-05 CIFAR100 VGG16 std 256/2 aa 0.0005 0.0 0.10 3395 1000 76.9 n
2025-05-05 CIFAR100 VGG16 std 256/2 aa 0.0005 0.0 0.10 9803 1000 77.7 n
2025-05-05 CIFAR100 VGG16 std 256/2 aa 0.0005 0.0 0.10 3791 1000 77.5 n
2025-05-05 CIFAR100 VGG16 std 256/2 aa 0.0010 0.0 0.10 3361 1000 77.6 n
2025-05-05 CIFAR100 VGG16 std 256/2 aa 0.0010 0.0 0.10 7170 1000 78.0 n
2025-05-05 CIFAR100 VGG16 std 256/2 aa 0.0010 0.0 0.10 2020 1000 77.7 n
2025-05-05 CIFAR100 VGG16 std 256/2 aa 0.0010 0.0 0.10 7935 1000 77.5 n
2025-02-11 CIFAR100 VGG16 std 256/8 aa 0.0004 0.0 1003 600 74.3 n
CIFAR100 VGG16 std 256 std 0.0 0.5 70.7 n
2025-05-02 CIFAR100 VGG16 std 512/1 aa 0.0005 0.0 0.20 997 923 76.9 n
2025-04-30 CIFAR100 VGG16 std 512/2 aa 0.0005 0.0 0.20 997 999 77.2 n
2025-04-30 CIFAR100 VGG16 std 512/2 aa 0.0005 0.0 0.20 998 896 77.0 n
2025-04-30 CIFAR100 VGG16 std 512/4 aa 0.0005 0.0 0.20 998 76.5 n
2025-04-29 CIFAR100 VGG16 std 512/8 aa 0.0005 0.0 0.20 998 74.8 n
CIFAR100 VGG16 std 512 aa 0.0 0.0 1001 72.9 n
2025-02-10 CIFAR100 VGG16 std 512/8 aa 0.0005 0.0 1001 73.4 n
CIFAR100 VGG16QCFS std 128 aa 0.0 0.0 1001 53.9 n
CIFAR100 VGG16QCFS std 256 aa 0.0 0.5 1001 72.0 n
2025-05-05 CIFAR100 VGG16QCFS std 256/2 aa 0.0010 0.0 0.10 7794 77.8 n
2025-05-05 CIFAR100 VGG16QCFS std 256/2 aa 0.0010 0.0 0.10 6862 77.7 n
2025-05-05 CIFAR100 VGG16QCFS std 256/2 aa 0.0010 0.0 0.10 4388 77.8 n
2025-05-05 CIFAR100 VGG16QCFS std 256/2 aa 0.0010 0.0 0.10 3086 77.8 n
2025-04-28 CIFAR100 VGG16QCFS std 256/4 aa 0.0005 0.0 0.05 999 75.9 n
2025-03-11 CIFAR100 VGG16QCFS std 256/8 aa 0.0005 0.0 0.05 999 984 76.1 n
2025-02-17 CIFAR100 VGG16QCFS std 256/8 aa 0.0005 0.0 0.05 1010 963 76.0 n
2025-02-17 CIFAR100 VGG16QCFS std 256/8 aa 0.0005 0.0 0.10 1010 966 76.4 n
2025-02-17 CIFAR100 VGG16QCFS std 256/8 aa 0.0005 0.0 0.20 1010 703 72.6 n
2025-04-29 CIFAR100 VGG16QCFS std 256/8 aa 0.0005 0.0 0.05 999 1000 75.8 n
2025-04-29 CIFAR100 VGG16QCFS std 256/8 aa 0.0005 0.0 0.10 999 724 74.0 n

Footnotes:

  • ConvNeXt trained with AdamW and OneCycleLR

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

点击 这是indexloc提供的php浏览器服务,不要输入任何密码和下载