@arizeai/phoenix-client 4.1.0 → 4.2.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@arizeai/phoenix-client",
-  "version": "4.1.0",
+  "version": "4.2.0",
   "description": "A client for the Phoenix API",
   "main": "dist/src/index.js",
   "module": "dist/esm/index.js",
@@ -67,7 +67,7 @@
     "tsx": "^4.19.3",
     "typescript": "^5.8.2",
     "vitest": "^2.1.9",
-    "@arizeai/phoenix-evals": "0.2.1"
+    "@arizeai/phoenix-evals": "0.2.2"
   },
   "dependencies": {
     "@arizeai/openinference-semantic-conventions": "^1.1.0",
@@ -98,7 +98,7 @@
     "prebuild": "pnpm run clean && pnpm run generate",
     "generate": "openapi-typescript --empty-objects-unknown=true --default-non-nullable=false ../../../schemas/openapi.json -o ./src/__generated__/api/v1.ts",
     "build": "tsc --build tsconfig.json tsconfig.esm.json && tsc-alias -p tsconfig.esm.json",
-    "postbuild": "echo '{\"type\": \"module\"}' > ./dist/esm/package.json && rimraf dist/test dist/examples",
+    "postbuild": "echo '{\"type\": \"module\"}' > ./dist/esm/package.json",
     "type:check": "tsc --noEmit",
     "test": "vitest --typecheck"
   }

package/src/datasets/createOrGetDataset.ts

@@ -0,0 +1,39 @@
+import { createClient } from "../client";
+import { CreateDatasetParams, createDataset } from "./createDataset";
+import { getDatasetInfoByName } from "./getDatasetInfoByName";
+
+export type CreateOrGetDatasetParams = CreateDatasetParams;
+
+export type CreateOrGetDatasetResponse = {
+  datasetId: string;
+};
+
+/**
+ * Given the parameters to create a dataset, this function will either
+ * retrieve an existing dataset by name or create a new one with the provided parameters.
+ *
+ * This is useful in cases where you would like to re-run a pipeline like:
+ * - ensure dataset exists
+ * - create a task
+ * - run experiment
+ * - evaluate experiment
+ * without having to create a new dataset each time.
+ */
+export async function createOrGetDataset({
+  name,
+  description,
+  examples,
+  client: _client,
+}: CreateOrGetDatasetParams): Promise<CreateOrGetDatasetResponse> {
+  const client = _client || createClient();
+  // start by fetching an existing dataset by name, catching any errors that occur
+  try {
+    const dataset = await getDatasetInfoByName({ datasetName: name, client });
+    return {
+      datasetId: dataset.id,
+    };
+  } catch {
+    // If the dataset doesn't exist, create it, falling back to the error handling inside createDataset
+    return await createDataset({ name, description, examples, client });
+  }
+}

package/src/datasets/index.ts

@@ -3,3 +3,4 @@ export * from "./getDataset";
 export * from "./getDatasetExamples";
 export * from "./appendDatasetExamples";
 export * from "./getDatasetInfo";
+export * from "./createOrGetDataset";

package/src/experiments/instrumentation.ts

@@ -20,6 +20,7 @@ export function createProvider({
   baseUrl,
   headers,
   useBatchSpanProcessor = true,
+  diagLogLevel,
 }: {
   projectName: string;
   headers: HeadersOptions;
@@ -32,8 +33,15 @@ export function createProvider({
    * The base URL of the Phoenix. Doesn't include the /v1/traces path.
    */
   baseUrl: string;
+  /**
+   * The diag log level to set for the built in DiagConsoleLogger instance.
+   * Omit to disable built in logging.
+   */
+  diagLogLevel?: DiagLogLevel;
 }) {
-  diag.setLogger(new DiagConsoleLogger(), DiagLogLevel.ERROR);
+  if (diagLogLevel) {
+    diag.setLogger(new DiagConsoleLogger(), diagLogLevel);
+  }
 
   const exporter = new OTLPTraceExporter({
     url: `${baseUrl}/v1/traces`,

package/src/experiments/runExperiment.ts

@@ -23,7 +23,12 @@ import { pluralize } from "../utils/pluralize";
 import { promisifyResult } from "../utils/promisifyResult";
 import { AnnotatorKind } from "../types/annotations";
 import { createProvider, createNoOpProvider } from "./instrumentation";
-import { SpanStatusCode, Tracer, trace } from "@opentelemetry/api";
+import {
+  type DiagLogLevel,
+  SpanStatusCode,
+  Tracer,
+  trace,
+} from "@opentelemetry/api";
 import {
   MimeType,
   OpenInferenceSpanKind,
@@ -111,6 +116,11 @@ export type RunExperimentParams = ClientFn & {
    * @default true
    */
   useBatchSpanProcessor?: boolean;
+  /**
+   * Log level to set for the default DiagConsoleLogger when tracing.
+   * Omit to disable default diag logging, or to bring your own.
+   */
+  diagLogLevel?: DiagLogLevel;
 };
 
 /**
@@ -160,6 +170,7 @@ export async function runExperiment({
   setGlobalTracerProvider = true,
   repetitions = 1,
   useBatchSpanProcessor = true,
+  diagLogLevel,
 }: RunExperimentParams): Promise<RanExperiment> {
   // Validation
   assert(
@@ -227,6 +238,7 @@ export async function runExperiment({
       baseUrl,
       headers: client.config.headers ?? {},
       useBatchSpanProcessor,
+      diagLogLevel,
     });
     // Register the provider
     if (setGlobalTracerProvider) {
@@ -298,6 +310,8 @@ export async function runExperiment({
     concurrency,
     dryRun,
     tracerProvider: provider,
+    diagLogLevel,
+    useBatchSpanProcessor,
   });
   ranExperiment.evaluationRuns = evaluationRuns;
 
@@ -468,6 +482,7 @@ export async function evaluateExperiment({
   setGlobalTracerProvider = true,
   useBatchSpanProcessor = true,
   tracerProvider: paramsTracerProvider,
+  diagLogLevel,
 }: {
   /**
    * The experiment to evaluate
@@ -502,6 +517,11 @@ export async function evaluateExperiment({
    * Intended as a pass-through from runExperiment
    */
   tracerProvider?: NodeTracerProvider | null;
+  /**
+   * Log level to set for the default DiagConsoleLogger when tracing.
+   * Omit to disable default diag logging, or to bring your own.
+   */
+  diagLogLevel?: DiagLogLevel;
 }): Promise<RanExperiment> {
   const isDryRun = typeof dryRun === "number" || dryRun === true;
   const client = _client ?? createClient();
@@ -521,6 +541,7 @@ export async function evaluateExperiment({
       baseUrl,
       headers: client.config.headers ?? {},
       useBatchSpanProcessor,
+      diagLogLevel,
     });
     if (setGlobalTracerProvider) {
       provider.register();