Accuracy

Available

image_classification.accuracy

The fraction of images whose top-scoring predicted class matches the ground-truth label.

Contract

Field	Value
version	1.0.0
metric_type	pointwise
scorer_contract	full_dataset
direction	higher_is_better
entrypoint	aip_scorers_image_classification.answered.accuracy
target_kind	None

Required columns

• image_id
• label
• predictions

Accepted schemas

[
  {
    "name": "gdi_image_v1",
    "task_types": [
      "classification"
    ]
  }
]

Methodology

1. Deterministic - no LLM. Uses numpy over the per-class score vectors.
2. For each image, pick the top-scoring class (argmax of the score vector).
3. Count the images where the predicted class equals the true label.
4. Divide by the number of images to get the 0-1 accuracy.

Score semantics

Scores range 0-1, the share of images classified correctly. A high score means most images were labelled correctly; a low score means many were misclassified. Higher is better.

Worked example

For 4 images where the top predicted class matches the truth on 3 of them -> 3/4 = 0.75.

Configuration schema

{
  "description": "Shared image-classification metric configuration.",
  "properties": {
    "inverted": {
      "default": false,
      "title": "Inverted",
      "type": "boolean",
      "x-aip-param-role": "scoring_metadata"
    },
    "max": {
      "default": 1.0,
      "title": "Max",
      "type": "number",
      "x-aip-param-role": "scoring_metadata"
    },
    "metric_name": {
      "default": "image_classification_metric",
      "title": "Metric Name",
      "type": "string",
      "x-aip-param-role": "scoring_metadata"
    },
    "min": {
      "default": 0.0,
      "title": "Min",
      "type": "number",
      "x-aip-param-role": "scoring_metadata"
    },
    "weight": {
      "default": 1.0,
      "title": "Weight",
      "type": "number",
      "x-aip-param-role": "scoring_metadata"
    }
  },
  "title": "InitializationConfig",
  "type": "object"
}

Execution

{
  "emits_metric_family": false,
  "function_name": null,
  "max_concurrency": null,
  "processing_kind": "cpu",
  "stream_batch_size": null,
  "timeout_seconds": 600
}

Complete manifest

accepts:
- name: gdi_image_v1
  task_types:
  - classification
config_schema:
  description: Shared image-classification metric configuration.
  properties:
    inverted:
      default: false
      title: Inverted
      type: boolean
      x-aip-param-role: scoring_metadata
    max:
      default: 1.0
      title: Max
      type: number
      x-aip-param-role: scoring_metadata
    metric_name:
      default: image_classification_metric
      title: Metric Name
      type: string
      x-aip-param-role: scoring_metadata
    min:
      default: 0.0
      title: Min
      type: number
      x-aip-param-role: scoring_metadata
    weight:
      default: 1.0
      title: Weight
      type: number
      x-aip-param-role: scoring_metadata
  title: InitializationConfig
  type: object
dependencies: []
description: Fraction of images where the top-scoring class matches the ground-truth
  label.
direction: higher_is_better
display_name: Accuracy
entrypoint: aip_scorers_image_classification.answered.accuracy
execution:
  emits_metric_family: false
  function_name: null
  max_concurrency: null
  processing_kind: cpu
  stream_batch_size: null
  timeout_seconds: 600
kind: metric
manifest_version: '1'
max_prompt_slots: 0
metric_metadata:
  methodology:
  - Deterministic - no LLM. Uses numpy over the per-class score vectors.
  - For each image, pick the top-scoring class (argmax of the score vector).
  - Count the images where the predicted class equals the true label.
  - Divide by the number of images to get the 0-1 accuracy.
  score_semantics: Scores range 0-1, the share of images classified correctly. A high
    score means most images were labelled correctly; a low score means many were misclassified.
    Higher is better.
  summary: The fraction of images whose top-scoring predicted class matches the ground-truth
    label.
  worked_example: For 4 images where the top predicted class matches the truth on
    3 of them -> 3/4 = 0.75.
metric_type: pointwise
name: image_classification.accuracy
partition_types: []
required_columns:
- image_id
- label
- predictions
required_kinds: []
scorer_contract: full_dataset
target_kind: null
unsupported_trace_shapes: []
version: 1.0.0