Class: Evaluation<R, E, M>

TypeScript SDK リファレンス

weave / Evaluation

一連の scorer と Datasets を含む評価（Evaluation）をセットアップします。

evaluation.evaluate(model) を呼び出すと、Datasets の行が Models に渡されます。このとき、Datasets の列名が model.predict の引数名と一致するようにマッピングされます。

その後、すべての scorer が呼び出され、その結果が Weave に保存されます。

Example

// 例題を Dataset に集約します
const dataset = new weave.Dataset({
  id: 'my-dataset',
  rows: [
    { question: 'What is the capital of France?', expected: 'Paris' },
    { question: 'Who wrote "To Kill a Mockingbird"?', expected: 'Harper Lee' },
    { question: 'What is the square root of 64?', expected: '8' },
  ],
});

// カスタムのスコアリング関数を定義します
const scoringFunction = weave.op(function isEqual({ modelOutput, datasetRow }) {
  return modelOutput == datasetRow.expected;
});

// 評価対象の関数を定義します
const model = weave.op(async function alwaysParisModel({ question }) {
  return 'Paris';
});

// 評価を開始します
const evaluation = new weave.Evaluation({
  id: 'my-evaluation',
  dataset: dataset,
  scorers: [scoringFunction],
});

const results = await evaluation.evaluate({ model });

​

Type parameters

名前	型
R	DatasetRow を継承
E	DatasetRow を継承
M	M

​

Hierarchy

• WeaveObject

  ↳ Evaluation

​

Table of contents

​

Constructors

• constructor

​

Properties

• __savedRef

​

Accessors

• description
• name

​

Methods

• evaluate
• predictAndScore
• saveAttrs

​

Constructors

​

constructor

• new Evaluation<R, E, M>(parameters): Evaluation<R, E, M>

​

Type parameters

名前	型
R	DatasetRow を継承
E	DatasetRow を継承
M	M

​

Parameters

名前	型
parameters	EvaluationParameters<R, E, M>

​

Returns

Evaluation<R, E, M>

​

Overrides

WeaveObject.constructor

​

Defined in

evaluation.ts:148

​

Properties

​

__savedRef

• Optional __savedRef: ObjectRef | Promise<ObjectRef>

​

Inherited from

WeaveObject.__savedRef

​

Defined in

weaveObject.ts:73

​

Accessors

​

description

• get description(): undefined | string

​

Returns

undefined | string

​

Inherited from

WeaveObject.description

​

Defined in

weaveObject.ts:100

---

​

name

• get name(): string

​

Returns

string

​

Inherited from

WeaveObject.name

​

Defined in

weaveObject.ts:96

​

Methods

​

evaluate

▸ evaluate(«destructured»): Promise<Record<string, any>>

​

Parameters

名前	型	デフォルト値
«destructured»	Object	undefined
› maxConcurrency?	number	5
› model	WeaveCallable<(…args: [{ datasetRow: R }]) => Promise<M>>	undefined
› nTrials?	number	1

​

Returns

Promise<Record<string, any>>

​

Defined in

evaluation.ts:163

---

​

predictAndScore

▸ predictAndScore(«destructured»): Promise<{ model_latency: number = modelLatency; model_output: any = modelOutput; model_success: boolean = !modelError; scores: { [key: string]: any; } }>

​

Parameters

名前	型
«destructured»	Object
› columnMapping?	ColumnMapping<R, E>
› example	R
› model	WeaveCallable<(…args: [{ datasetRow: E }]) => Promise<M>>

​

Returns

Promise<{ model_latency: number = modelLatency; model_output: any = modelOutput; model_success: boolean = !modelError; scores: { [key: string]: any; } }>

​

Defined in

evaluation.ts:231

---

​

saveAttrs

▸ saveAttrs(): Object

​

Returns

Object

​

Inherited from

WeaveObject.saveAttrs

​

Defined in

weaveObject.ts:77