Add experiment link in the SDK (#353)

* Add experiment link

* Add mock for get_experiment_url method

* Fix linter issues

* Update example

sdks/python/examples/evaluation_example.py

@@ -1,9 +1,6 @@
 from typing import Dict, Any
 
-from opik.evaluation.metrics import (
-    IsJson,
-    Hallucination,
-)
+from opik.evaluation.metrics import IsJson, Hallucination
 from opik.evaluation import evaluate
 from opik import Opik, DatasetItem, track
 from opik.integrations.openai import track_openai
@@ -15,8 +12,6 @@
 
 openai_client = track_openai(openai.OpenAI())
 
-# contains_hello = Contains(searched_value="hello", name="ContainsHello")
-# contains_bye = Contains(searched_value="bye", name="ContainsBye")
 is_json = IsJson()
 hallucination = Hallucination()
 
@@ -63,6 +58,7 @@ def llm_task(item: DatasetItem) -> Dict[str, Any]:
         "output": response.choices[0].message.content,
         "input": item.input["message"],
         "context": item.input["context"],
+        "reference": "test",
     }
 
 

sdks/python/src/opik/evaluation/evaluator.py

@@ -75,6 +75,9 @@ def evaluate(
         dataset_name=dataset.name,
         experiment_config=experiment_config,
     )
+
+    report.display_experiment_link(dataset.name, experiment.id)
+
     experiment_items = [
         experiment_item.ExperimentItem(
             dataset_item_id=result.test_case.dataset_item_id,

sdks/python/src/opik/evaluation/report.py

@@ -2,6 +2,7 @@
 from typing import List, Dict, Tuple
 from collections import defaultdict
 from . import test_result
+from .. import url_helpers
 
 
 def _format_time(seconds: float) -> str:
@@ -81,3 +82,13 @@ def display_experiment_results(
     console_container = console.Console()
     console_container.print(panel_content)
     console_container.print("Uploading results to Opik ... ")
+
+
+def display_experiment_link(dataset_name: str, experiment_id: str) -> None:
+    console_container = console.Console()
+    experiment_url = url_helpers.get_experiment_url(
+        dataset_name=dataset_name, experiment_id=experiment_id
+    )
+    console_container.print(
+        f"View the results [link={experiment_url}]in your Opik dashboard[/link]."
+    )

sdks/python/src/opik/url_helpers.py

@@ -1,4 +1,5 @@
 import opik.config
+import opik.api_objects.opik_client
 
 
 def get_ui_url() -> str:
@@ -8,6 +9,21 @@ def get_ui_url() -> str:
     return opik_url_override.rstrip("/api")
 
 
+def get_experiment_url(dataset_name: str, experiment_id: str) -> str:
+    client = opik.api_objects.opik_client.get_client_cached()
+
+    # Get dataset id from name
+    dataset = client._rest_client.datasets.get_dataset_by_identifier(
+        dataset_name=dataset_name
+    )
+    dataset_id = dataset.id
+
+    config = opik.config.OpikConfig()
+    ui_url = get_ui_url()
+
+    return f'{ui_url}/{config.workspace}/experiments/{dataset_id}/compare?experiments=%5B"{experiment_id}"%5D'
+
+
 def get_projects_url() -> str:
     config = opik.config.OpikConfig()
     ui_url = get_ui_url()

sdks/python/tests/unit/evaluation/test_evaluate.py

@@ -3,7 +3,7 @@
 
 from opik.api_objects.dataset import dataset_item
 from opik.api_objects import opik_client
-from opik import evaluation, exceptions
+from opik import evaluation, exceptions, url_helpers
 from opik.evaluation import metrics
 from ...testlib import backend_emulator_message_processor, ANY_BUT_NONE, assert_equal
 from ...testlib.models import (
@@ -56,6 +56,9 @@ def say_task(dataset_item: dataset_item.DatasetItem):
     mock_create_experiment = mock.Mock()
     mock_create_experiment.return_value = mock_experiment
 
+    mock_get_experiment_url = mock.Mock()
+    mock_get_experiment_url.return_value = "any_url"
+
     with mock.patch.object(
         opik_client.Opik, "create_experiment", mock_create_experiment
     ):
@@ -64,13 +67,16 @@ def say_task(dataset_item: dataset_item.DatasetItem):
             "construct_online_streamer",
             mock_construct_online_streamer,
         ):
-            evaluation.evaluate(
-                dataset=mock_dataset,
-                task=say_task,
-                experiment_name="the-experiment-name",
-                scoring_metrics=[metrics.Equals()],
-                task_threads=1,
-            )
+            with mock.patch.object(
+                url_helpers, "get_experiment_url", mock_get_experiment_url
+            ):
+                evaluation.evaluate(
+                    dataset=mock_dataset,
+                    task=say_task,
+                    experiment_name="the-experiment-name",
+                    scoring_metrics=[metrics.Equals()],
+                    task_threads=1,
+                )
 
     mock_create_experiment.assert_called_once_with(
         dataset_name="the-dataset-name",