feat: Display button to view evaluation results in an Ipython environment.

PiperOrigin-RevId: 713454093

Author: vertex-sdk-bot

google/cloud/aiplatform/utils/_ipython_utils.py

@@ -17,6 +17,7 @@
 
 import sys
 import typing
+import urllib
 from uuid import uuid4
 from typing import Optional
 
@@ -266,3 +267,13 @@ def display_browse_prebuilt_metrics_button() -> None:
         "https://cloud.google.com/vertex-ai/generative-ai/docs/models/metrics-templates"
     )
     display_link("Browse pre-built metrics", uri, "list")
+
+
+def display_gen_ai_evaluation_results_button(gcs_file_path: str) -> None:
+    """Function to generate a link bound to the Gen AI evaluation run."""
+    if not is_ipython_available():
+        return
+
+    gcs_file_path = urllib.parse.quote(gcs_file_path)
+    uri = f"https://console.cloud.google.com/storage/browser/_details/{gcs_file_path}"
+    display_link("View evaluation results", uri, "bar_chart")

tests/unit/vertexai/test_evaluation.py

@@ -558,13 +558,9 @@ def mock_experiment_tracker():
 
 
 @pytest.fixture
-def mock_storage_blob_upload_from_filename():
-    with mock.patch(
-        "google.cloud.storage.Blob.upload_from_filename"
-    ) as mock_blob_upload_from_filename, mock.patch(
-        "google.cloud.storage.Bucket.exists", return_value=True
-    ):
-        yield mock_blob_upload_from_filename
+def mock_storage_blob_from_string():
+    with mock.patch("google.cloud.storage.Blob.from_string") as mock_blob_from_string:
+        yield mock_blob_from_string
 
 
 @pytest.mark.usefixtures("google_auth_mock")
@@ -1948,8 +1944,29 @@ def test_pairtwise_metric_prompt_template_with_default_values(self):
             == _EXPECTED_PAIRWISE_PROMPT_TEMPLATE_WITH_DEFAULT_VALUES.strip()
         )
 
-    def test_upload_results(self, mock_storage_blob_upload_from_filename):
-        evaluation.utils.upload_evaluation_results(
-            _TEST_CSV, _TEST_BUCKET, _TEST_FILE_NAME
+    def test_upload_results(self, mock_storage_blob_from_string):
+        with mock.patch("json.dump") as mock_json_dump:
+            evaluation.utils.upload_evaluation_results(
+                MOCK_EVAL_RESULT,
+                _TEST_BUCKET,
+                _TEST_FILE_NAME,
+                "candidate_model",
+                "baseline_model",
+            )
+
+        mock_storage_blob_from_string.assert_any_call(
+            uri="gs://test-bucket/test-file-name/test-file-name.csv",
+            client=mock.ANY,
+        )
+        mock_storage_blob_from_string.assert_any_call(
+            uri="gs://test-bucket/test-file-name/summary_metrics.json",
+            client=mock.ANY,
+        )
+        mock_json_dump.assert_called_once_with(
+            {
+                "summary_metrics": MOCK_EVAL_RESULT.summary_metrics,
+                "candidate_model_name": "candidate_model",
+                "baseline_model_name": "baseline_model",
+            },
+            mock.ANY,
         )
-        assert mock_storage_blob_upload_from_filename.called_once_with(_TEST_CSV)

vertexai/evaluation/eval_task.py

@@ -464,8 +464,29 @@ def evaluate(
                 evaluation_service_qps=evaluation_service_qps,
                 retry_timeout=retry_timeout,
             )
+
+        candidate_model_name = None
+        if isinstance(model, generative_models.GenerativeModel):
+            candidate_model_name = model._model_name
+
+        baseline_model_name = None
+        pairwise_metrics = [
+            metric
+            for metric in self.metrics
+            if isinstance(metric, pairwise_metric.PairwiseMetric)
+        ]
+        if pairwise_metrics:
+            # All pairwise metrics should have the same baseline model.
+            baseline_model = pairwise_metrics[0].baseline_model
+            if isinstance(baseline_model, generative_models.GenerativeModel):
+                baseline_model_name = baseline_model._model_name
+
         utils.upload_evaluation_results(
-            eval_result.metrics_table, self.output_uri_prefix, output_file_name
+            eval_result,
+            self.output_uri_prefix,
+            output_file_name,
+            candidate_model_name,
+            baseline_model_name,
         )
         return eval_result
 

vertexai/evaluation/utils.py

@@ -17,21 +17,24 @@
 
 import functools
 import io
+import json
 import os
 import tempfile
 import threading
 import time
-from typing import Any, Dict, Optional, TYPE_CHECKING, Union, Callable, Literal
+from typing import Any, Callable, Dict, Literal, Optional, TYPE_CHECKING, Union
 
 from google.cloud import bigquery
 from google.cloud import storage
 from google.cloud.aiplatform import base
 from google.cloud.aiplatform import compat
 from google.cloud.aiplatform import initializer
 from google.cloud.aiplatform import utils
+from google.cloud.aiplatform.utils import _ipython_utils
 from google.cloud.aiplatform_v1.services import (
     evaluation_service as gapic_evaluation_services,
 )
+from vertexai.evaluation import _base as eval_base
 
 
 if TYPE_CHECKING:
@@ -275,32 +278,77 @@ def _upload_pandas_df_to_gcs(
                 " Please provide a valid GCS path with `jsonl` or `csv` suffix."
             )
 
-        storage_client = storage.Client(
-            project=initializer.global_config.project,
-            credentials=initializer.global_config.credentials,
-        )
-        storage.Blob.from_string(
-            uri=upload_gcs_path, client=storage_client
-        ).upload_from_filename(filename=local_dataset_path)
+        _upload_file_to_gcs(upload_gcs_path, local_dataset_path)
+
+
+def _upload_evaluation_summary_to_gcs(
+    summary_metrics: Dict[str, float],
+    upload_gcs_path: str,
+    candidate_model_name: Optional[str] = None,
+    baseline_model_name: Optional[str] = None,
+):
+    """Uploads the evaluation summary to a GCS bucket."""
+    summary = {
+        "summary_metrics": summary_metrics,
+    }
+    if candidate_model_name:
+        summary["candidate_model_name"] = candidate_model_name
+    if baseline_model_name:
+        summary["baseline_model_name"] = baseline_model_name
+
+    with tempfile.TemporaryDirectory() as temp_dir:
+        local_summary_path = os.path.join(temp_dir, "summary_metrics.json")
+        json.dump(summary, open(local_summary_path, "w"))
+        _upload_file_to_gcs(upload_gcs_path, local_summary_path)
+
+
+def _upload_file_to_gcs(upload_gcs_path: str, filename: str):
+    storage_client = storage.Client(
+        project=initializer.global_config.project,
+        credentials=initializer.global_config.credentials,
+    )
+    storage.Blob.from_string(
+        uri=upload_gcs_path, client=storage_client
+    ).upload_from_filename(filename)
 
 
 def upload_evaluation_results(
-    dataset: "pd.DataFrame", destination_uri_prefix: str, file_name: str
+    eval_result: eval_base.EvalResult,
+    destination_uri_prefix: str,
+    file_name: str,
+    candidate_model_name: Optional[str] = None,
+    baseline_model_name: Optional[str] = None,
 ) -> None:
     """Uploads eval results to GCS destination.
 
     Args:
-        dataset: Pandas dataframe to upload.
+        eval_result: Eval results to upload.
         destination_uri_prefix: GCS folder to store the data.
-        file_name: File name to store the data.
+        file_name: File name to store the metrics table.
+        candidate_model_name: Optional. Candidate model name.
+        baseline_model_name: Optional. Baseline model name.
     """
     if not destination_uri_prefix:
         return
+    if eval_result.metrics_table is None:
+        return
     if destination_uri_prefix.startswith(_GCS_PREFIX):
-        _, extension = os.path.splitext(file_name)
+        base_name, extension = os.path.splitext(file_name)
         file_type = extension.lower()[1:]
-        output_path = destination_uri_prefix + "/" + file_name
-        _upload_pandas_df_to_gcs(dataset, output_path, file_type)
+        output_folder = destination_uri_prefix + "/" + base_name
+        metrics_table_path = output_folder + "/" + file_name
+        _upload_pandas_df_to_gcs(
+            eval_result.metrics_table, metrics_table_path, file_type
+        )
+        _upload_evaluation_summary_to_gcs(
+            eval_result.summary_metrics,
+            output_folder + "/summary_metrics.json",
+            candidate_model_name,
+            baseline_model_name,
+        )
+        _ipython_utils.display_gen_ai_evaluation_results_button(
+            metrics_table_path.split(_GCS_PREFIX)[1]
+        )
     else:
         raise ValueError(
             f"Unsupported destination URI: {destination_uri_prefix}."