awslabs / sagemaker-debugger-rulesconfig Goto Github PK

With Python 3.12.2, when I import the ,sagemaker sdk, an import of this package is triggered and I see

***/site-packages/smdebug_rulesconfig/actions/utils.py:6: SyntaxWarning: invalid escape sequence '\w'
  EMAIL_ADDRESS_REGEX = "^[a-z0-9]+[@]\w+[.]\w{2,3}$"

and a few others. For proper escaping, I think the easiest is to use raw strings, like EMAIL_ADDRESS_REGEX = r"^[a-z0-9]+[@]\w+[.]\w{2,3}$", which should be fairly backwards compatible.

Also note that in the future, this warning will become an error:

Changed in version 3.12: Octal escapes with value larger than 0o377 produce a SyntaxWarning. In a future Python version they will be eventually a SyntaxError.

Note this may conflict with the open PR #37.

In validate_email_address, it checks if EMAIL_ADDRESS_REGEX and the email address in the argument match, but the result is false for email addresses like [email protected]. Can you please fix this so that the result is true for email addresses with the above pattern?

I am working on adding code snippets of rule configuration to the Debugger dev guide.
I am referring to the following source codes in order to see what default collections are taken by each rule.

• https://github.com/awslabs/sagemaker-debugger-rulesconfig/blob/master/smdebug_rulesconfig/rule_config_jsons/ruleConfigs.json
• https://github.com/awslabs/sagemaker-debugger-rules/tree/master/smdebug_rules

In the ruleConfigs.json, some rules have default collections, but some are missing them.
I suggest to add a default tensor collection to those rules that currently don't have it.

For example, in order for the XGBoost TreeDepth rule to work, we need to add the tree collection by default as following:

  "TreeDepth": {
    "DebugRuleConfiguration": {
      "RuleConfigurationName": "TreeDepth",
      "RuleParameters": {
        "rule_to_invoke": "TreeDepth"
      }
    },
    "CollectionConfigurations": [
      {
        "CollectionName": "tree",
        "CollectionParameters": {
          "save_interval": "500"
        }
      }
    ]
  },

On pypi, will you guys be able to release Source file instead of only Wheel file?

Thanks.

I was reviewing the new notebook regarding StalledTrainingRule.
This is a new built-in 1p rule, and I wonder why the way to call this rule is different from other existing built-in rules.
I cannot call it simply by using the following method:

from sagemaker.debugger import Rule, rule_configs

stalled_training_job_rule = [Rule.sagemaker(rule_configs.stalled_training_rule())]

And the rule also cannot be imported by running the following code in a JupyterLab notebook cell:

from smdebug_rulesconfig import stalled_training_rule

Error message:

ImportErrorTraceback (most recent call last)
<ipython-input-22-1af5fa9dfa3f> in <module>()
----> 1 from smdebug_rulesconfig import stalled_training_rule

ImportError: cannot import name stalled_training_rule

I made sure to install smdebug-rulesconfig-0.1.4, and this problem still persists.

It would be nice if the way to use this stalled_training_rule is as same as using the existing built-in rules.

For the setup.py, we include the all the packages:

packages=setuptools.find_packages()

Actually, the tests folder will not be used in production, is that possible to exclude this tests folder to avoid the tests folder installed?