Add ORACLE to LSST Science Pipeline

broker/cloud_run/lsst/oracle/Dockerfile

@@ -0,0 +1,21 @@
+# Use the official lightweight Python image.
+# https://hub.docker.com/_/python
+FROM python:3.12-slim
+
+# Allow statements and log messages to immediately appear in the Knative logs
+ENV PYTHONUNBUFFERED True
+
+# Copy local code to the container image.
+ENV APP_HOME /app
+WORKDIR $APP_HOME
+COPY . ./
+
+# Install production dependencies.
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Run the web service on container startup. Here we use the gunicorn
+# webserver, with one worker process and 8 threads.
+# For environments with multiple CPU cores, increase the number of workers
+# to be equal to the cores available.
+# Timeout is set to 0 to disable the timeouts of the workers to allow Cloud Run to handle instance scaling.
+CMD exec gunicorn --bind :$PORT --workers 1 --threads 8 --timeout 0 main:app

broker/cloud_run/lsst/oracle/cloudbuild.yaml

@@ -0,0 +1,29 @@
+# https://cloud.google.com/build/docs/deploying-builds/deploy-cloud-run
+# containerize the module and deploy it to Cloud Run
+steps:
+# Build the image
+- name: 'gcr.io/cloud-builders/docker'
+  args: ['build', '-t', '${_REGION}-docker.pkg.dev/${PROJECT_ID}/${_REPOSITORY}/${_MODULE_IMAGE_NAME}', '.']
+# Push the image to Artifact Registry
+- name: 'gcr.io/cloud-builders/docker'
+  args: ['push', '${_REGION}-docker.pkg.dev/${PROJECT_ID}/${_REPOSITORY}/${_MODULE_IMAGE_NAME}']
+# Deploy image to Cloud Run
+- name: 'gcr.io/google.com/cloudsdktool/cloud-sdk'
+  entrypoint: gcloud
+  args: ['run', 'deploy', '${_MODULE_NAME}', '--image', '${_REGION}-docker.pkg.dev/${PROJECT_ID}/${_REPOSITORY}/${_MODULE_IMAGE_NAME}', '--region', '${_REGION}', '--set-env-vars', '${_ENV_VARS}']
+images:
+- '${_REGION}-docker.pkg.dev/${PROJECT_ID}/${_REPOSITORY}/${_MODULE_IMAGE_NAME}'
+substitutions:
+    _SURVEY: 'lsst'
+    _TESTID: 'testid'
+    _MODULE_NAME: '${_SURVEY}-classify_with_ORACLE-${_TESTID}'
+    _MODULE_IMAGE_NAME: 'gcr.io/${PROJECT_ID}/${_REPOSITORY}/${_MODULE_NAME}'
+    _REPOSITORY: 'cloud-run-services'
+    # cloud functions automatically sets the projectid env var using the name "GCP_PROJECT"
+    # use the same name here for consistency
+    # [TODO] PROJECT_ID is set in setup.sh. this is confusing and we should revisit the decision.
+    # i (Raen) think i didn't make it a substitution because i didn't want to set a default for it.
+    _ENV_VARS: 'GCP_PROJECT=${PROJECT_ID},SURVEY=${_SURVEY},TESTID=${_TESTID}'
+    _REGION: 'us-central1'
+options:
+    dynamic_substitutions: true

broker/cloud_run/lsst/oracle/deploy.sh

@@ -0,0 +1,89 @@
+#! /bin/bash
+# Deploys or deletes broker Cloud Run service
+# This script will not delete Cloud Run services that are in production
+
+# "False" uses production resources
+# any other string will be appended to the names of all resources
+testid="${1:-test}"
+# "True" tearsdown/deletes resources, else setup
+teardown="${2:-False}"
+# name of the survey this broker instance will ingest
+survey="${3:-lsst}"
+region="${4:-us-central1}"
+# get the environment variable
+PROJECT_ID=$GOOGLE_CLOUD_PROJECT
+
+MODULE_NAME="oracle"  # lower case required by cloud run
+ROUTE_RUN="/"  # url route that will trigger main.run()
+
+define_GCP_resources() {
+    local base_name="$1"
+    local separator="${2:--}"
+    local testid_suffix=""
+
+    if [ "$testid" != "False" ] && [ -n "$testid" ]; then
+        testid_suffix="${separator}${testid}"
+    fi
+    echo "${base_name}${testid_suffix}"
+}
+
+#--- GCP resources used in this script
+artifact_registry_repo=$(define_GCP_resources "${survey}-cloud-run-services")
+bq_dataset=$(define_GCP_resources "${survey}" "_")
+bq_table="ORACLE"
+cr_module_name=$(define_GCP_resources "${survey}-${MODULE_NAME}")  # lower case required by cloud run
+ps_input_subscrip=$(define_GCP_resources "${survey}-${MODULE_NAME}") # pub/sub subscription used to trigger cloud run module
+ps_output_topic=$(define_GCP_resources "${survey}-${MODULE_NAME}")
+ps_trigger_topic=$(define_GCP_resources "${survey}-lite")
+runinvoker_svcact="cloud-run-invoker@${PROJECT_ID}.iam.gserviceaccount.com"
+# topics and subscriptions involved in writing data to BigQuery
+ps_bigquery_subscription=$(define_GCP_resources "${survey}-${MODULE_NAME}-bigquery-import")
+ps_deadletter_topic=$(define_GCP_resources "${survey}-deadletter")
+
+if [ "${teardown}" = "True" ]; then
+    # ensure that we do not teardown production resources
+    if [ "${testid}" != "False" ]; then
+        echo
+        echo "Deleting resources for ${MODULE_NAME} module..."
+        gcloud pubsub topics delete "${ps_output_topic}"
+        gcloud pubsub subscriptions delete "${ps_bigquery_subscription}"
+        gcloud pubsub subscriptions delete "${ps_input_subscrip}"
+        gcloud run services delete "${cr_module_name}" --region "${region}"
+    fi
+else
+    echo "Configuring Pub/Sub resources..."
+    gcloud pubsub topics create "${ps_output_topic}"
+    gcloud pubsub subscriptions create "${ps_bigquery_subscription}" \
+        --topic="${ps_output_topic}" \
+        --bigquery-table="${PROJECT_ID}:${bq_dataset}.${bq_table}" \
+        --use-table-schema \
+        --drop-unknown-fields \
+        --dead-letter-topic="${ps_deadletter_topic}" \
+        --max-delivery-attempts=5 \
+        --dead-letter-topic-project="${PROJECT_ID}"
+    # set IAM policies on public Pub/Sub resources
+    if [ "$testid" = "False" ]; then
+        user="allUsers"
+        roleid="projects/${GOOGLE_CLOUD_PROJECT}/roles/userPublic"
+        gcloud pubsub topics add-iam-policy-binding "${ps_output_topic}" --member="${user}" --role="${roleid}"
+    fi
+
+    #--- Deploy Cloud Run service
+    echo
+    echo "Creating container image for ${MODULE_NAME} module and deploying to Cloud Run..."
+    moduledir="."  # deploys what's in our current directory
+    config="${moduledir}/cloudbuild.yaml"
+    url=$(gcloud builds submit --config="${config}" \
+        --substitutions="_SURVEY=${survey},_TESTID=${testid},_MODULE_NAME=${cr_module_name},_REPOSITORY=${artifact_registry_repo}" \
+        "${moduledir}" | sed -n 's/^Step #2: Service URL: \(.*\)$/\1/p')
+    echo
+    echo "Creating trigger subscription for ${MODULE_NAME} Cloud Run service..."
+    gcloud pubsub subscriptions create "${ps_input_subscrip}" \
+        --topic "${ps_trigger_topic}" \
+        --topic-project "${PROJECT_ID}" \
+        --ack-deadline=600 \
+        --push-endpoint="${url}${ROUTE_RUN}" \
+        --push-auth-service-account="${runinvoker_svcact}" \
+        --dead-letter-topic="${ps_deadletter_topic}" \
+        --max-delivery-attempts=5
+fi

broker/cloud_run/lsst/oracle/main.py

@@ -0,0 +1,152 @@
+#!/usr/bin/env python3
+# -*- coding: UTF-8 -*-
+"""Classify an alert using ORACLE (Shah et al. 2025).
+
+This code is intended to be containerized and deployed to Google Cloud Run.
+Once deployed, individual alerts in the "trigger" stream will be delivered to the container as HTTP requests.
+"""
+
+import os
+from pathlib import Path
+import flask  # Manage the HTTP request containing the alert
+import pittgoogle  # Manipulate the alert and interact with cloud resources
+
+import pandas as pd
+from astroOracle.pretrained_models import ORACLE_lite
+
+import google.cloud.logging
+
+# [FIXME] Make this helpful or else delete it.
+# Connect the python logger to the google cloud logger.
+# By default, this captures INFO level and above.
+# pittgoogle uses the python logger.
+# We don't currently use the python logger directly in this script, but we could.
+google.cloud.logging.Client().setup_logging()
+
+# These environment variables are defined when running the deploy.sh script.
+PROJECT_ID = os.getenv("GCP_PROJECT")
+TESTID = os.getenv("TESTID")
+SURVEY = os.getenv("SURVEY")
+
+# provenance variables
+MODULE_NAME = "ORACLE"
+MODULE_VERSION = 0.1
+
+# classifier variables
+model_dir_name = "./"
+model_file_name = "best_no_md_model.h5"
+MODEL_PATH = Path(__file__).resolve().parent / model_dir_name / model_file_name
+
+# Variables for incoming data
+# A url route is used in setup.sh when the trigger subscription is created.
+# It is possible to define multiple routes in a single module and trigger them using different subscriptions.
+ROUTE_RUN = "/"  # HTTP route that will trigger run(). Must match setup.sh
+
+# Variables for outgoing data
+HTTP_204 = 204  # HTTP code: Success
+HTTP_400 = 400  # HTTP code: Bad Request
+
+# GCP resources used in this module
+# pittgoogle will construct the full resource names from the MODULE_NAME, SURVEY, and TESTID
+TOPIC = pittgoogle.Topic.from_cloud(
+    MODULE_NAME, survey=SURVEY, testid=TESTID, projectid=PROJECT_ID
+)
+
+app = flask.Flask(__name__)
+
+@app.route(ROUTE_RUN, methods=["POST"])
+def run():
+    """Classify the alert; publish and store results.
+
+    This module is intended to be deployed as a Cloud Run service. It will operate as an HTTP endpoint
+    triggered by Pub/Sub messages. This function will be called once for every message sent to this route.
+    It should accept the incoming HTTP request and return a response.
+
+    Returns
+    -------
+    response : tuple(str, int)
+        Tuple containing the response body (string) and HTTP status code (int). Flask will convert the
+        tuple into a proper HTTP response. Note that the response is a status message for the web server
+        and should not contain the classification results.
+    """
+    # extract the envelope from the request that triggered the endpoint
+    # this contains a single Pub/Sub message with the alert to be processed
+    envelope = flask.request.get_json()
+
+    # unpack the alert. raises a `BadRequest` if the envelope does not contain a valid message
+    try:
+        alert_lite = pittgoogle.Alert.from_cloud_run(envelope, "default")
+    except pittgoogle.exceptions.BadRequest as exc:
+        return str(exc), HTTP_400
+
+    model_lite = ORACLE_lite(MODEL_PATH)
+    oracle_classification = model_lite.predict([_format_for_classifier(alert_lite)]).to_dict(orient='records')[0]
+
+    level_1_class = _most_likely_class(oracle_classification, ['Transient', 'Variable'])
+    level_2_class = _most_likely_class(oracle_classification, ['SN', 'Fast', 'Long',    # Transient
+                                                               'Periodic', 'AGN'])      # Variable
+    leaf_class = _most_likely_class(oracle_classification, ['SNIa', 'SNIb/c', 'SNIax', 'SNI91bg', 'SNII',   # SN
+                                                            'KN', 'Dwarf Novae', 'uLens', 'M-dwarf Flare',  # Fast
+                                                            'SLSN', 'TDE', 'ILOT', 'CART', 'PISN',          # Long
+                                                            'Cepheid', 'RR Lyrae', 'Delta Scuti', 'EB',     # Periodic
+                                                            'AGN'])                                         # AGN
+
+    # publish
+    outpt_dict = {
+        "output": oracle_classification,
+        "predicted_level_1": level_1_class[0],
+        "predicted_level_1_prob": level_1_class[1],
+        "predicted_level_2": level_2_class[0],
+        "predicted_level_2_prob": level_2_class[1],
+        "predicted_leaf": leaf_class[0],
+        "predicted_leaf_prob": leaf_class[1]
+    }
+
+    TOPIC.publish(
+        pittgoogle.Alert.from_dict(
+            payload={'alert_lite': alert_lite.dict['alert_lite'], 'ORACLE': outpt_dict},
+            attributes={
+                **alert_lite.attributes,
+                'pg_oracle_class': outpt_dict['predicted_leaf'],
+            },
+            schema_name='default',
+        )
+    )
+
+    return "", HTTP_204
+
+def _y_to_Y(band):
+    if band == 'y':
+        return 'Y'
+    return band
+
+# this could use improvement
+def _get_photflag(flux, flux_err):
+    if flux_err > 5*abs(flux):
+        return 1024
+    elif abs(flux) < max(100, flux_err):
+        return 0
+    else:
+        return 4096
+
+def _format_for_classifier(alert_lite: pittgoogle.Alert) -> pd.DataFrame:
+    """Create a DataFrame for input to ORACLE."""
+    alert_lite_dict = alert_lite.dict['alert_lite']
+    source_dicts = [alert_lite_dict['diaSource']] + alert_lite_dict['prvDiaSources'] + alert_lite_dict['prvDiaForcedSources']
+
+    MJD_min = min([source_dict['midpointMjdTai'] for source_dict in source_dicts])
+    df = pd.DataFrame({'MJD': [source_dict['midpointMjdTai'] - MJD_min for source_dict in source_dicts],    # start dates at 0
+                       'BAND': [_y_to_Y(source_dict['band']) for source_dict in source_dicts],               # ORACLE wants Y band to be capital
+                       'FLUXCAL': [source_dict['psfFlux'] for source_dict in source_dicts],
+                       'FLUXCALERR': [source_dict['psfFluxErr'] for source_dict in source_dicts],
+                       'PHOTFLAG': [_get_photflag(source_dict['psfFlux'], source_dict['psfFluxErr']) for source_dict in source_dicts]})
+    return df.sort_values(by=['MJD'])
+
+def _most_likely_class(probability_dict: dict, keys: list) -> tuple[str, float]:
+    max_val = 0
+    max_class = None
+    for key in keys:
+        if max_val < probability_dict[key]:
+            max_val = probability_dict[key]
+            max_class = key
+    return (max_class, max_val)

broker/cloud_run/lsst/oracle/requirements.txt

@@ -0,0 +1,15 @@
+# As explained here
+# https://cloud.google.com/functions/docs/writing/specifying-dependencies-python
+# dependencies for a Cloud Function must be specified in a `requirements.txt`
+# file (or packaged with the function) in the same directory as `main.py`
+
+google-cloud-logging
+pittgoogle-client>=0.3.14
+git+https://github.com/uiucsn/Astro-ORACLE.git
+
+# for Cloud Run
+# https://cloud.google.com/run/docs/quickstarts/build-and-deploy/deploy-python-service
+# pinned following quickstart example. [TODO] consider un-pinning
+Flask
+gunicorn
+Werkzeug