ahangchen / torch_base Goto Github PK

View Code? Open in Web Editor NEW

545.0 5.0 105.0 40 KB

Quickly bring up your PyTorch project(a skeleton)

Python 100.00%

pytorch computer-vision deep-learning machine-learning

torch_base's Introduction

TorchBase

This is a project about how to do nn experiments methodically with pytorch.

More instruction: 如何让炼丹更有条理

torch_base's People

Contributors

Stargazers

Watchers

Forkers

wangqiang9 wzylife15 chenjingtao-jinan lighterswang zhiyumeng jinghangli chenbingxiayu storyandwine tragedyn xiaohuangyang chh13502 lxb6476 gaonbnb victor-future woruoweikunlun ipanda555 zjzhang123 wanghuancheng ntubenquan helloworldiamjerry tonylibing kakuibeyond caicandong wallufo breaker505 harryingit3 shixiangnan u-zzd lxb1226 yongsun1998 gongminghu diavro m1andy arslan-z jackleiaaaaaa royess hjj194 kennyng-19 luoxubo eltheren fengshi-cherish zoroastrian-nova iamfaith stanmae liq77 yuanhub9 chinesehuang tbhacker hapuhxy xxn-1n wenze-tu chunqizhang hh-hh19 chendi77 chimesz xinjianouyang blessyyyu cyp0630 yuanpei0208 zznew rocklqy little1tow longlongvip betbetliu xdjiangkai yenchiang zhangjiahuan17 fyj-codingidiot huangwei0102 jiafengttang spyfighting yifeilin0226 ming053l pekahzero jevonyang zhi-fangzhang hekenye nasame ssscrl kungfu-crab sunmmyuan iq-scm sdu-huanhq weleven9991 yotta2020 utopia124 zxthu huaiyuechusan kepeng2019 weayer404 westcraneli simon007-heiyewuxing zhemglee transformerzz suojifeng xxwwsz shixishi flyingbird93 boreas618 wen-98

torch_base's Issues

Typo here

torch_base/utils/logger.py

Line 14 in 5d32e95

self.metrics[name] = []

should be self.metrics[name] = [value]

a little mistake in list_dataset.py

list_dataset.py：
line 16：self.label_paths = [infos[1] for info in infos]-->self.label_paths = [info[1] for info in infos]

torch.manual_seed(args.seed)

在train.py文件中第21行torch.manual_seed(args.seed)，
但是在options.py文件中好像并没有设置args.seed...

Some places confused me in `options.py`

Hi there!

torch_base/options.py

Lines 63 to 64 in a57aab4

 ext = os.path.basename(args.load_model_path).split('.')[0] 

 model_dir = args.load_model_path.replace(ext, '')

According to these two lines, if we set arguments as default, (load_model_path='checkpoints/base_model_pref/0.pth', save_prefix='pref', val_list='/data/dataset1/list/base/val.txt'),

then the result_dir shows like checkpoints/base_model_pref/.pth/base_val_pref.

I wonder if there are any mistakes or do you just want to save test results under the .pth folder?

Thanks for your contribution~

A little confusion -- train.py

I got a bit confused here,
line 33 in train.py,
should be 'self.model.load_state_dict(torch.load(args.load_model_path))' or 'self.model.load_state_dict(torch.load(args.load_model_path).state_dict())'?

[Help Wanted] DistributedDataParallel (DDP) instead of DataParallel (DP)

Hi, very tidy and efficient base code!
Do you have a plan to implement DistributedDataParallel (DDP) instead of DataParallel (DP) for training with multiple gpus?