This is relevant when you logging Reports to the Platform. Tags help you associate the Report with a specific model / prompt version, period, etc.

​
Add timestamp

Each Report run has a single timestamp. By default, Evidently will assign the datetime.now() using the Report computation time based on the user time zone.

You can also add your own timestamp. Pass it to the run method:

from datetime import datetime

my_eval_4 = report.run(eval_data_1,
                       eval_data_2,
                       timestamp=datetime(2024, 1, 29))

Since you can assign arbitrary timestamps, you can log Reports asynchronously or with a delay (for example, after you receive ground truth) and assign it to the specific period.

​
Add tags and metadata

You can add tags and metadata to Reports. This is useful for search and data filtering. By adding Tags, you can also visualize data only from a specific subset of Reports on a monitoring Panel.

Examples of when to use tags include:

  • Identify a specific evaluation run or experiment (by model or prompt version, test scenario, etc).

  • Identify app or model status: production/shadow, champion/challenger, or A/B model versions.

  • Identify the Report by geography, use case, segment, role, etc.

  • Identify the reference dataset window (for example, if you compare distribution drift week-by-week and month-by-month and log two different Reports).

  • You want to tag individual Reports that have a special role, e.g., a datasheet card, a model card, etc.

Custom tags. Pass any custom Tags as a list:

report = Report([
    ClassificationPreset()
],
tags=["groupA", "shadow"])

Custom metadata. Pass metadata as a Python dictionary in key:value pairs:

report = Report([
    ClassificationPreset()
],
metadata = {
	"deployment": "shadow",
	"status": "production",
	})

Default metadata. Use built-in metadata fields model_id, reference_id, batch_size, dataset_id:

report = Report([
    ClassificationPreset()
],
  model_id="model_id",
	reference_id="reference_id",
	batch_size="batch_size",
	dataset_id="dataset_id"
)
Output formatsIntroduction