docs(people-and-planet-ai): align wildlife notebook with updated clou…

people-and-planet-ai/image-classification/README.ipynb

@@ -67,7 +67,6 @@
     "    </a>\n",
     "  </td>\n",
     "</table>\n",
-    "<!-- Legacy redirect aliases: https://console.cloud.google.com/vertex-ai/colab/import/https:%2F%2Fraw.githubusercontent.com%2FGoogleCloudPlatform%2Fpython-docs-samples%2Fmain%2Fpeople-and-planet-ai%2Fimage-classification%2FREADME.ipynb and https://console.cloud.google.com/vertex-ai/workbench/deploy-notebook?download_url=https://raw.githubusercontent.com/GoogleCloudPlatform/python-docs-samples/main/people-and-planet-ai/image-classification/README.ipynb -->\n",
     "\n",
     "<br clear=\"left\"/>\n",
     "\n",
@@ -81,7 +80,13 @@
     "gcloud compute networks subnets create default --project=example-project-id --network=default --region=us-central1 --range=10.128.0.0/20 --enable-private-ip-google-access\n",
     "```\n",
     "\n",
-    "</details>"
+    "</details>\n",
+    "\n",
+    "---\n",
+    "\n",
+    "| Author |\n",
+    "| --- |\n",
+    "| [Google Cloud](https://github.com/GoogleCloudPlatform) |"
    ],
    "metadata": {
     "id": "1LUt29D0MSeu"
@@ -115,7 +120,33 @@
     "   Use [`classify_wildlife_image.py`](classify_wildlife_image.py) to inspect camera trap photos, see how taxonomic family (`taxon_family`) resolves look-alike species, and pass `habitat_region` context to geo-fence predictions.\n",
     "4. **Generate multimodal embeddings for rare-species search**:\n",
     "   Use [`embed_wildlife_image.py`](embed_wildlife_image.py) and **Gemini Embedding 2** (`gemini-embedding-2`) to turn a single reference photo of an endangered species into a 768-dimensional vector embedding for similarity search in BigQuery.\n",
-    "5. **Clean up resources** to avoid ongoing storage costs."
+    "5. **Clean up resources** to avoid ongoing storage costs.\n",
+    "\n",
+    "### Objectives\n",
+    "\n",
+    "In this tutorial, you learn how to:\n",
+    "\n",
+    "- Ingest and filter LILA WCS Camera Traps metadata into BigQuery using Apache Beam on Dataflow (`create_images_metadata_table.py`).\n",
+    "- Run class-balanced sampling (`Sample.FixedSizePerKey`) and batch multimodal inference (`RunInference` with `gemini-3.8-flash`) directly against `gs://public-datasets-lila/wcs-unzipped/` (`run_wildlife_inference.py`).\n",
+    "- Classify individual camera trap photos into structured ecological observations with taxonomic hierarchy (`taxon_family`) and habitat geo-fencing (`classify_wildlife_image.py`).\n",
+    "- Generate 768-dimensional multimodal vector embeddings with `gemini-embedding-2` and run BigQuery `VECTOR_SEARCH` for rare-species discovery (`embed_wildlife_image.py`).\n",
+    "\n",
+    "### Costs\n",
+    "\n",
+    "This tutorial uses billable components of Google Cloud:\n",
+    "\n",
+    "- [Dataflow](https://cloud.google.com/dataflow/pricing)\n",
+    "- [BigQuery](https://cloud.google.com/bigquery/pricing)\n",
+    "- [Vertex AI / Gemini API](https://cloud.google.com/vertex-ai/generative-ai/pricing)\n",
+    "- [Cloud Storage](https://cloud.google.com/storage/pricing)\n",
+    "\n",
+    "Use the [Google Cloud Pricing Calculator](https://cloud.google.com/products/calculator) to generate a cost estimate based on your projected usage.\n",
+    "\n",
+    "### Prerequisites\n",
+    "\n",
+    "- A Google Cloud project with billing enabled.\n",
+    "- The Dataflow (`dataflow.googleapis.com`), Vertex AI (`aiplatform.googleapis.com`), and BigQuery (`bigquery.googleapis.com`) APIs enabled.\n",
+    "- IAM roles: Dataflow Admin (`roles/dataflow.admin`), BigQuery Admin (`roles/bigquery.admin`), Storage Admin (`roles/storage.admin`), and Vertex AI User (`roles/aiplatform.user`)."
    ],
    "metadata": {
     "id": "jXDRG8QK2BZD"
@@ -154,19 +185,19 @@
     "    # Install libffi-dev for the Dataflow pipeline in ephemeral Colab runtimes.\n",
     "    !apt-get -qq install libffi-dev\n",
     "\n",
-    "    # Sparse-checkout only people-and-planet-ai/image-classification into an isolated directory.\n",
-    "    REPO_DIR = \"ppai-image-classification\"\n",
-    "    SAMPLE_SUBPATH = \"people-and-planet-ai/image-classification\"\n",
-    "    BRANCH = \"main\"\n",
+    "# Sparse-checkout only people-and-planet-ai/image-classification into an isolated directory.\n",
+    "REPO_DIR = \"ppai-image-classification\"\n",
+    "SAMPLE_SUBPATH = \"people-and-planet-ai/image-classification\"\n",
+    "BRANCH = \"main\"\n",
     "\n",
-    "    if not os.getcwd().endswith(SAMPLE_SUBPATH):\n",
-    "        if not os.path.exists(REPO_DIR):\n",
-    "            !git clone --filter=blob:none --no-checkout --depth 1 --sparse -b {BRANCH} https://github.com/GoogleCloudPlatform/python-docs-samples.git {REPO_DIR}\n",
-    "            !git -C {REPO_DIR} sparse-checkout set {SAMPLE_SUBPATH}\n",
-    "        %cd {REPO_DIR}\n",
-    "        !git fetch --depth 1 origin {BRANCH}\n",
-    "        !git -c advice.detachedHead=false checkout -f FETCH_HEAD\n",
-    "        %cd {SAMPLE_SUBPATH}\n",
+    "if not os.getcwd().endswith(SAMPLE_SUBPATH):\n",
+    "    if not os.path.exists(REPO_DIR):\n",
+    "        !git clone --filter=blob:none --no-checkout --depth 1 --sparse -b {BRANCH} https://github.com/GoogleCloudPlatform/python-docs-samples.git {REPO_DIR}\n",
+    "        !git -C {REPO_DIR} sparse-checkout set {SAMPLE_SUBPATH}\n",
+    "    %cd {REPO_DIR}\n",
+    "    !git fetch --depth 1 origin {BRANCH}\n",
+    "    !git -c advice.detachedHead=false checkout -f FETCH_HEAD\n",
+    "    %cd {SAMPLE_SUBPATH}\n",
     "\n",
     "if os.getcwd() not in sys.path:\n",
     "    sys.path.insert(0, os.getcwd())\n",
@@ -251,11 +282,13 @@
    "cell_type": "code",
    "execution_count": null,
    "source": [
+    "# fmt: off\n",
     "PROJECT_ID = \"example-project-id\"  # @param {type: \"string\", placeholder: \"example-project-id\", isTemplate: true}\n",
+    "LOCATION = \"us-central1\"  # @param {type: \"string\"}\n",
+    "# fmt: on\n",
+    "\n",
     "if not PROJECT_ID or PROJECT_ID == \"example-project-id\":\n",
     "    PROJECT_ID = os.environ.get(\"GOOGLE_CLOUD_PROJECT\", \"example-project-id\")\n",
-    "\n",
-    "LOCATION = \"us-central1\"  # @param {type: \"string\"}\n",
     "if not LOCATION:\n",
     "    LOCATION = os.environ.get(\"GOOGLE_CLOUD_LOCATION\", \"us-central1\")\n",
     "\n",
@@ -780,12 +813,50 @@
    "execution_count": null,
    "outputs": []
   },
+  {
+   "cell_type": "markdown",
+   "source": [
+    "### Validate target state\n",
+    "\n",
+    "Verify that all 6 gallery camera trap frames were embedded into 768-dimensional vectors and written to `embeddings.jsonl`:"
+   ],
+   "metadata": {
+    "id": "validate-target-state-md"
+   }
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "with open(\"embeddings.jsonl\") as f:\n",
+    "    written_rows = [json.loads(line) for line in f if line.strip()]\n",
+    "\n",
+    "assert len(written_rows) == len(gallery), (\n",
+    "    f\"Expected {len(gallery)} rows in embeddings.jsonl, got {len(written_rows)}\"\n",
+    ")\n",
+    "for i, record in enumerate(written_rows):\n",
+    "    assert record[\"image_gcs_uri\"] == gallery[i][0], (\n",
+    "        f\"Row {i} URI mismatch: {record['image_gcs_uri']} != {gallery[i][0]}\"\n",
+    "    )\n",
+    "    assert record[\"species_scientific_name\"] == gallery[i][1], (\n",
+    "        f\"Row {i} species mismatch: {record['species_scientific_name']} != {gallery[i][1]}\"\n",
+    "    )\n",
+    "    assert len(record[\"embedding\"]) == 768, (\n",
+    "        f\"Row {i} expected 768-d embedding, got {len(record['embedding'])}\"\n",
+    "    )\n",
+    "print(\"Target state validated: 6 camera trap embeddings (768 dimensions) verified in embeddings.jsonl.\")"
+   ],
+   "metadata": {
+    "id": "validate-target-state-cell"
+   },
+   "execution_count": null,
+   "outputs": []
+  },
   {
    "cell_type": "markdown",
    "source": [
     "## Cleaning up\n",
     "\n",
-    "To avoid incurring ongoing storage charges for the resources created in this notebook, run the following cell to delete the BigQuery dataset and the temporary Dataflow staging files in Cloud Storage:"
+    "To avoid incurring ongoing storage charges for the resources created in this notebook, set `delete_resources = True` and run the following cell to delete the BigQuery dataset and the temporary Dataflow staging files in Cloud Storage:"
    ],
    "metadata": {
     "id": "MGl_1UyiGVl6"
@@ -794,7 +865,9 @@
   {
    "cell_type": "code",
    "source": [
-    "delete_resources = True  # @param {type:\"boolean\"}\n",
+    "# fmt: off\n",
+    "delete_resources = False  # @param {type:\"boolean\"}\n",
+    "# fmt: on\n",
     "\n",
     "if delete_resources or os.getenv(\"IS_TESTING\"):\n",
     "    # Delete the BigQuery dataset (including the metadata, predictions, and embeddings tables).\n",
@@ -804,6 +877,7 @@
     "    !gcloud storage rm -r \"gs://{BUCKET_NAME}/{CLOUD_STORAGE_DIRECTORY}\""
    ],
    "metadata": {
+    "cellView": "form",
     "id": "8418262e"
    },
    "execution_count": null,

people-and-planet-ai/image-classification/e2e_test.py

@@ -117,7 +117,12 @@ def test_extract_metadata_fn_filters_invalid_categories() -> None:
     mock_response = mock.MagicMock()
     mock_response.content = zip_buffer.getvalue()
 
-    with mock.patch("requests.get", return_value=mock_response):
+    with mock.patch.object(
+        create_images_metadata_table.requests,
+        "get",
+        return_value=mock_response,
+        autospec=True,
+    ):
         fn = create_images_metadata_table._ExtractMetadataFn()
         results = list(fn.process(create_images_metadata_table.METADATA_URL))