ClearML/clearml-docs
1
0
Fork 0
mirror of https://github.com/clearml/clearml-docs synced 2025-02-07 13:21:46 +00:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Add setting random seed for task (#318)

Browse Source
This commit is contained in:
pollfly 2022-08-30 15:36:40 +03:00 committed by GitHub
parent c4049bbb11
commit ff9f897629
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 14 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
docs/clearml_sdk/task_sdk.md
View File

@ -13,6 +13,11 @@ populate it with:
* Argparse arguments (default and specific to the current execution)
* Reports to Tensorboard & Matplotlib and model checkpoints.
:::tip Ensuring Reproducibility
To ensure every run will provide the same results, ClearML controls the deterministic behaviors of the `tensorflow`,
`pytorch`, and `random` packages by setting a fixed initial seed. See [Setting Random Seed](#setting-random-seed).
:::
:::note
ClearML object (e.g. task, project) names are required to be at least 3 characters long
:::
@ -373,6 +378,15 @@ Upload the execution data that the Task captured offline to the ClearML Server u
Both options will upload the Task's full execution details and outputs and return a link to the Task's results page on
the ClearML Server.
### Setting Random Seed
To ensure task reproducibility, ClearML controls the deterministic behaviors of the `tensorflow`, `pytorch`, and `random`
packages by setting a fixed initial seed.
ClearML uses `1337` as the default initial seed. To set a different value for your task, use the [`Task.set_random_seed`](../references/sdk/task.md#taskset_random_seed)
class method and provide the new seed value, before initializing the task.
You can disable the deterministic behavior entirely by passing `Task.set_random_seed(None)`.
## Artifacts
Artifacts are the output files created by a task. ClearML uploads and logs these products so they can later be easily
accessed, modified, and used.