[Question]modify the initial config

[GarryJAY502]

❓ Question

Hi,
We had a question about the config file.
How should I modify the initial epoch batch learnrate and other parameters during training.
which file
Thanks!

[partha-ghosh]

Hi @GarryJAY502 ,
If the config.yaml you can find :

defaults:
    - _self_
    - train: retinaunet_hnm_v002

The configuration works by inheriting configuration from retinaunet_hnm_v002.yaml in the train folder. The same way, you can find the configurations retinaunet_hnm_v002.yaml is inheriting from other config files. The trainer-specific configuration is in the trainer_cfg folder, where you can update the learning rate, number of epochs etc.

You can also update the parameters through the command line itself. For example, to update the warm_iterations to 5000 and initial_lr to 0.001, you have to pass -o trainer_cfg.warm_iterations=5000 trainer_cfg.initial_lr=0.001. This way you can update other parameters as well.

Best,
Partha

[GarryJAY502]

你好@GarryJAY502， 如果config.yaml您能找到：

defaults:
    - _self_
    - train: retinaunet_hnm_v002

配置通过从文件夹retinaunet_hnm_v002.yaml中继承配置来工作train。同样，您可以发现配置retinaunet_hnm_v002.yaml是从其他配置文件继承的。训练器特定的配置位于trainer_cfg文件夹中，您可以在其中更新学习率、时期数等。

您还可以通过命令行本身更新参数。例如，要将更新warm_iterations为 5000 和initial_lr0.001，您必须传递-o trainer_cfg.warm_iterations=5000 trainer_cfg.initial_lr=0.001。这样您也可以更新其他参数。

最好的， 帕塔

It works，thanks ,Partha