Datasets¶

Datasets store evaluation data for testing and benchmarking your LLM applications.

Creating Datasets¶

// Create a simple dataset
dataset, _ := client.CreateDataset(ctx, "my-dataset")

// Create with options
dataset, _ := client.CreateDataset(ctx, "evaluation-data",
    opik.WithDatasetDescription("Test data for Q&A evaluation"),
    opik.WithDatasetTags("qa", "evaluation", "v1"),
)

Inserting Items¶

Single Item¶

dataset.InsertItem(ctx, map[string]any{
    "input":    "What is the capital of France?",
    "expected": "Paris",
    "category": "geography",
})

Multiple Items¶

items := []map[string]any{
    {
        "input":    "What is 2+2?",
        "expected": "4",
        "category": "math",
    },
    {
        "input":    "What is 3+3?",
        "expected": "6",
        "category": "math",
    },
    {
        "input":    "What color is the sky?",
        "expected": "Blue",
        "category": "general",
    },
}

dataset.InsertItems(ctx, items)

Retrieving Items¶

// Get items with pagination
items, _ := dataset.GetItems(ctx, 1, 100) // page 1, 100 items per page

for _, item := range items {
    fmt.Printf("Input: %v\n", item.Data["input"])
    fmt.Printf("Expected: %v\n", item.Data["expected"])
}

Listing Datasets¶

// List all datasets
datasets, _ := client.ListDatasets(ctx, 1, 100)

for _, ds := range datasets {
    fmt.Printf("Dataset: %s (ID: %s)\n", ds.Name, ds.ID)
}

Getting a Dataset by Name¶

dataset, _ := client.GetDatasetByName(ctx, "my-dataset")

fmt.Printf("Name: %s\n", dataset.Name)
fmt.Printf("Description: %s\n", dataset.Description)

Deleting Datasets¶

// Delete by dataset object
dataset.Delete(ctx)

Dataset Item Structure¶

Each dataset item can contain any fields you need:

Common Field	Description
input	The input/prompt to evaluate
expected	Expected/ground truth output
context	Additional context for evaluation
metadata	Any additional metadata

Example: Building an Evaluation Dataset¶

func buildEvaluationDataset(ctx context.Context, client *opik.Client) (*opik.Dataset, error) {
    // Create dataset
    dataset, err := client.CreateDataset(ctx, "qa-evaluation-v1",
        opik.WithDatasetDescription("Q&A evaluation dataset"),
        opik.WithDatasetTags("qa", "production"),
    )
    if err != nil {
        return nil, err
    }

    // Load test cases from file or database
    testCases := loadTestCases()

    // Convert to dataset items
    items := make([]map[string]any, len(testCases))
    for i, tc := range testCases {
        items[i] = map[string]any{
            "input":    tc.Question,
            "expected": tc.Answer,
            "context":  tc.Context,
            "category": tc.Category,
        }
    }

    // Bulk insert
    err = dataset.InsertItems(ctx, items)
    if err != nil {
        return nil, err
    }

    return dataset, nil
}

Using Datasets with Experiments¶

See Experiments for how to run evaluations against datasets.