Implementation:Truera Trulens Feedback On Selectors

Knowledge Sources	TruLens TruLens Docs
Domains	LLM_Evaluation, Observability
Last Updated	2026-02-14 08:00 GMT

Overview

Concrete tool for defining composable evaluation metrics with selector-based data binding provided by the trulens-core library.

Description

The Feedback class (deprecated alias for Metric) provides a composable API for specifying evaluation functions and binding their parameters to specific data from application traces. The class supports chaining methods like .on_input(), .on_output(), .on_input_output(), and .on() to bind function parameters to trace selectors. An optional aggregation function combines multiple evaluation results.

Note: The Feedback class is deprecated in favor of the Metric class. The API is identical; only the class name and imp parameter name (now implementation) have changed.

Usage

Import and use after configuring a feedback provider. Create Feedback/Metric instances when defining evaluation criteria for application traces. Use the RAG Triad pattern (answer relevance, context relevance, groundedness) as a baseline for RAG applications.

Code Reference

Source Location

Repository: trulens
File: src/core/trulens/core/feedback/feedback.py (Feedback class, L37-108)
File: src/core/trulens/core/metric/metric.py (Metric class with on_input, on_output, on methods, L420-803)

Signature

class Feedback(Metric):
    def __init__(
        self,
        imp: Optional[Callable] = None,
        agg: Optional[Callable] = None,
        examples: Optional[List[Tuple]] = None,
        criteria: Optional[str] = None,
        additional_instructions: Optional[str] = None,
        min_score_val: Optional[int] = 0,
        max_score_val: Optional[int] = 3,
        temperature: Optional[float] = 0.0,
        groundedness_configs: Optional[GroundednessConfigs] = None,
        enable_trace_compression: Optional[bool] = None,
        **kwargs,
    ):
        """
        Args:
            imp: The feedback function to execute (DEPRECATED: use implementation).
            agg: Aggregator function for combining multiple results.
            criteria: Criteria for the feedback evaluation.
            additional_instructions: Custom instructions for the LLM judge.
            min_score_val: Minimum score value (default: 0).
            max_score_val: Maximum score value (default: 3).
            temperature: Temperature for LLM-based feedback (default: 0.0).
        """

    # Selector chaining methods (inherited from Metric):
    def on_input(self, arg: Optional[str] = None) -> Metric: ...
    def on_output(self, arg: Optional[str] = None) -> Metric: ...
    def on_input_output(self) -> Metric: ...
    def on(self, *args, **kwargs) -> Metric: ...

Import

# Deprecated import:
from trulens.core import Feedback

# Recommended import:
from trulens.core import Metric

I/O Contract

Inputs

Name	Type	Required	Description
imp / implementation	Callable	Yes	Feedback function from a provider (e.g., provider.context_relevance)
agg	Callable	No	Aggregation function (e.g., numpy.mean)
criteria	str	No	Custom evaluation criteria text
additional_instructions	str	No	Extra instructions for the LLM judge
min_score_val	int	No	Minimum score (default: 0)
max_score_val	int	No	Maximum score (default: 3)
temperature	float	No	LLM temperature (default: 0.0)

Outputs

Name	Type	Description
return	Feedback/Metric	Configured evaluation metric with selector bindings, ready for use with app wrappers

Usage Examples

RAG Triad Feedback Functions

from trulens.core import Feedback
from trulens.providers.openai import OpenAI
import numpy as np

provider = OpenAI()

# Answer relevance: input -> output
f_answer_relevance = Feedback(
    provider.relevance
).on_input_output()

# Context relevance: input -> each context chunk
f_context_relevance = (
    Feedback(provider.context_relevance)
    .on_input()
    .on(TruChain.select_context())
    .aggregate(np.mean)
)

# Groundedness: context -> output
f_groundedness = (
    Feedback(
        provider.groundedness_measure_with_cot_reasons,
        name="Groundedness"
    )
    .on(TruChain.select_context().collect())
    .on_output()
)

Using the New Metric API

from trulens.core import Metric

metric = Metric(
    implementation=provider.relevance
).on_input_output()

Related Pages

Implements Principle

Principle:Truera_Trulens_Feedback_Function_Definition

Requires Environment

Environment:Truera_Trulens_Python_Core_Environment

Uses Heuristic

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment