Maybe it makes sense to limit or change number of parameters which participate in name generation or allow users to use their own naming scheme. https://github.com/allegroai/clearml/blob/24464b7c1019f7a7b3149ecb80a379c5f82337a0/clearml/automation/optimization.py#L720
I run model training for weeks, where I train hundreds of models with different hyperparameters. Therefore, I need automatic naming of newly created tasks. Maybe there are some examples of how to set custom task naming scheme?
AgitatedDove14
Stackoverflow says about the size of the file name, but here the problem is with the size of the folder name. Not sure if the file and folder limits are the same. The error that occurs says that An error occurred (XMinioInvalidObjectName) when calling the CompleteMultipartUpload operation: Object name contains unsupported characters.
. At first I thought I was using the wrong characters in the task title, such as backslash, colon or space. But I managed to upload models with the task name including all these symbols when I used fewer override parameters in my hypeparameters optimization task. So I assumed that the problem is in the length of the folder. Here is an example.
Maybe you should make naming_function as public variable in SearchStrategy class or allow changing it in HyperParameterOptimizer class?
MinuteGiraffe30 , Hi ! 🙂
What if you try to manually create such a folder?
Thanks MinuteGiraffe30 , fix will be pushed later today
Here is an issue https://github.com/allegroai/clearml/issues/516 .
Maybe you should make
naming_function
as public variable in
SearchStrategy
class or allow changing it in
HyperParameterOptimizer
class?
I like this idea, let's do that
Just making sure, you hit the 1024 character limit on S3 path?
If this is the case we should also fix the "artifact naming" to take that into account (it already does and has a limit, see here:
https://github.com/allegroai/clearml/blob/24464b7c1019f7a7b3149ecb80a379c5f82337a0/clearml/backend_interface/metrics/events.py#L351
Should we also limit the entire full path?
I came up with a workaround to solve this problem, but since I run tasks remotely, where not my modified clearml code is used, this problem will remain until you fix it. Should I duplicate this issue in the github?