feat: save proof prediction for all proofs in a new table

.env

@@ -40,3 +40,16 @@ POSTGRES_EXPOSE=127.0.0.1:5432
 ENVIRONMENT=net
 
 GUNICORN_WORKERS=1
+
+# We use special `host.docker.internal` to access the localhost (=your laptop) from
+# the docker container
+# It works because we added the special `host.docker.internal:host-gateway`
+# host in dev.yml for all services
+# Triton is the ML inference server used at Open Food Facts
+TRITON_URI=host.docker.internal:5004
+
+# By default, don't enable ML predictions, as we don't necessarily have a Triton
+# server running.
+# During local development, to enable ML predictions, set this to True and make sure
+# you have Triton running on port 5004.
+ENABLE_ML_PREDICTIONS=False

.github/workflows/generate-docs.yml

@@ -55,7 +55,7 @@
             # we only deploy on push to main
             if: |
               github.event_name == 'push' && github.event.ref == 'refs/heads/main'
-            uses: JamesIves/github-pages-deploy-action@v4.6.8
+            uses: JamesIves/github-pages-deploy-action@v4.7.1
             with:
               GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
               BRANCH: gh-pages # The branch the action should deploy to.

Makefile

@@ -160,9 +160,8 @@ cli: guard-args
 	${DOCKER_COMPOSE} run --rm --no-deps api python3 manage.py ${args}
 
 
-# TODO: migrate to Django
-add-db-revision: guard-message
-	${DOCKER_COMPOSE} run --rm --no-deps api alembic revision --autogenerate -m "${message}"
+makemigrations: guard-args
+	${DOCKER_COMPOSE} run --rm --no-deps api python3 manage.py makemigrations ${args}
 
 #---------#
 # Cleanup #

config/settings.py

@@ -289,3 +289,9 @@
 # ------------------------------------------------------------------------------
 
 GOOGLE_GEMINI_API_KEY = os.getenv("GOOGLE_GEMINI_API_KEY")
+
+# Triton Inference Server (ML)
+# ------------------------------------------------------------------------------
+
+TRITON_URI = os.getenv("TRITON_URI", "localhost:5004")
+ENABLE_ML_PREDICTIONS = os.getenv("ENABLE_ML_PREDICTIONS") == "True"

docker-compose.yml

@@ -16,6 +16,8 @@ x-api-common: &api-common
     - LOG_LEVEL
     - GOOGLE_CLOUD_VISION_API_KEY
     - GOOGLE_GEMINI_API_KEY
+    - TRITON_URI
+    - ENABLE_ML_PREDICTIONS
   networks:
     - default
 

docker/dev.yml

@@ -22,6 +22,9 @@ x-api-base: &api-base
     # mount tests
     - ./tests:/opt/open-prices/tests
     - ./manage.py:/opt/open-prices/manage.py
+  # Allow the container to access the host's network, for Triton server to be accessible
+  extra_hosts:
+    - "host.docker.internal:host-gateway"
 
 services:
   api:

open_prices/proofs/constants.py

@@ -10,3 +10,16 @@
 TYPE_SINGLE_SHOP_LIST = [TYPE_PRICE_TAG, TYPE_RECEIPT, TYPE_SHOP_IMPORT]
 TYPE_SHOPPING_SESSION_LIST = [TYPE_RECEIPT, TYPE_GDPR_REQUEST]
 TYPE_MULTIPLE_SHOP_LIST = [TYPE_GDPR_REQUEST]
+
+PROOF_PREDICTION_OBJECT_DETECTION_TYPE = "OBJECT_DETECTION"
+PROOF_PREDICTION_CLASSIFICATION_TYPE = "CLASSIFICATION"
+PROOF_PREDICTION_RECEIPT_EXTRACTION_TYPE = "RECEIPT_EXTRACTION"
+PROOF_PREDICTION_PRICE_TAG_EXTRACTION_TYPE = "PRICE_TAG_EXTRACTION"
+PROOF_PREDICTION_LIST = [
+    PROOF_PREDICTION_OBJECT_DETECTION_TYPE,
+    PROOF_PREDICTION_CLASSIFICATION_TYPE,
+    PROOF_PREDICTION_RECEIPT_EXTRACTION_TYPE,
+    PROOF_PREDICTION_PRICE_TAG_EXTRACTION_TYPE,
+]
+
+PROOF_TYPE_CHOICES = [(key, key) for key in PROOF_PREDICTION_LIST]

open_prices/proofs/migrations/0006_add_proof_prediction_table.py

@@ -0,0 +1,101 @@
+# Generated by Django 5.1 on 2024-12-01 17:41
+
+import django.db.models.deletion
+import django.utils.timezone
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+    dependencies = [
+        ("proofs", "0005_proof_receipt_price_count_proof_receipt_price_total"),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name="ProofPrediction",
+            fields=[
+                (
+                    "id",
+                    models.BigAutoField(
+                        auto_created=True,
+                        primary_key=True,
+                        serialize=False,
+                        verbose_name="ID",
+                    ),
+                ),
+                (
+                    "type",
+                    models.CharField(
+                        choices=[
+                            ("OBJECT_DETECTION", "OBJECT_DETECTION"),
+                            ("CLASSIFICATION", "CLASSIFICATION"),
+                            ("RECEIPT_EXTRACTION", "RECEIPT_EXTRACTION"),
+                            ("PRICE_TAG_EXTRACTION", "PRICE_TAG_EXTRACTION"),
+                        ],
+                        max_length=20,
+                        verbose_name="The type of the prediction",
+                    ),
+                ),
+                (
+                    "model_name",
+                    models.CharField(
+                        max_length=30,
+                        verbose_name="The name of the model that generated the prediction",
+                    ),
+                ),
+                (
+                    "model_version",
+                    models.CharField(
+                        max_length=30,
+                        verbose_name="The specific version of the model that generated the prediction",
+                    ),
+                ),
+                (
+                    "created",
+                    models.DateTimeField(
+                        default=django.utils.timezone.now,
+                        verbose_name="When the prediction was created in DB",
+                    ),
+                ),
+                (
+                    "data",
+                    models.JSONField(
+                        blank=True,
+                        null=True,
+                        verbose_name="a dict representing the data of the prediction. This field is model-specific.",
+                    ),
+                ),
+                (
+                    "value",
+                    models.CharField(
+                        blank=True,
+                        max_length=30,
+                        null=True,
+                        verbose_name="The predicted value, only for classification models, null otherwise.",
+                    ),
+                ),
+                (
+                    "max_confidence",
+                    models.FloatField(
+                        blank=True,
+                        null=True,
+                        verbose_name="The maximum confidence of the prediction, may be null for some models.For object detection models, this is the confidence of the most confident object.For classification models, this is the confidence of the predicted class.",
+                    ),
+                ),
+                (
+                    "proof",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="predictions",
+                        to="proofs.proof",
+                        verbose_name="The proof this prediction belongs to",
+                    ),
+                ),
+            ],
+            options={
+                "verbose_name": "Proof Prediction",
+                "verbose_name_plural": "Proof Predictions",
+                "db_table": "proof_predictions",
+            },
+        ),
+    ]