利用地点数据洞见创建自定义位置得分

CREATE OR REPLACE TABLE `your_project.your_dataset.apartment_listings`
(
  id INT64,
  name STRING,
  location GEOGRAPHY
);

INSERT INTO `your_project.your_dataset.apartment_listings` VALUES
  (1, 'The Downtowner', ST_GEOGPOINT(-74.0077, 40.7093)),
  (2, 'Suburban Oasis', ST_GEOGPOINT(-73.9825, 40.7507)),
  (3, 'Riverside Lofts', ST_GEOGPOINT(-73.9470, 40.8079))
  -- More rows can be added here
  . . . ;

-- This Common Table Expression (CTE) will hold the raw counts for each score component.
WITH insight_counts AS (
  SELECT WITH AGGREGATION_THRESHOLD -- Correctly includes the mandatory aggregation threshold
    apartments.id,
    apartments.name,
    COUNTIF(places.primary_type = 'park') AS park_count,
    COUNTIF(places.primary_type = 'museum') AS museum_count,
    COUNTIF(places.primary_type = 'restaurant' AND places.good_for_children = TRUE) AS family_restaurant_count,
    COUNTIF(places.primary_type IN ('veterinary_care', 'pet_store')) AS pet_service_count,
    COUNTIF(places.allows_dogs = TRUE) AS dog_friendly_place_count
  FROM
    `your_project.your_dataset.apartment_listings` AS apartments
  LEFT JOIN
    `your-project.places_insights___us.places` AS places -- Corrected table name for the US dataset
    ON ST_DWITHIN(apartments.location, places.point, 800) -- Find places within 800 meters
  GROUP BY
    apartments.id, apartments.name
)
SELECT * FROM insight_counts;

WITH
  -- CTE 1: Count nearby amenities of interest for each apartment listing.
  insight_counts AS (
    SELECT WITH AGGREGATION_THRESHOLD
      apartments.id,
      apartments.name,
      COUNTIF(places.primary_type = 'park') AS park_count,
      COUNTIF(places.primary_type = 'museum') AS museum_count,
      COUNTIF(places.primary_type = 'restaurant' AND places.good_for_children = TRUE) AS family_restaurant_count,
      COUNTIF(places.primary_type IN ('veterinary_care', 'pet_store')) AS pet_service_count,
      COUNTIF(places.allows_dogs = TRUE) AS dog_friendly_place_count
    FROM
      `your_project.your_dataset.apartment_listings` AS apartments
    LEFT JOIN
      `your-project.places_insights___us.places` AS places
      ON ST_DWITHIN(apartments.location, places.point, 800)
    GROUP BY
      apartments.id,
      apartments.name
  ),
  -- CTE 2: Apply custom weighting to the amenity counts to generate raw scores.
  raw_scores AS (
    SELECT
      id,
      name,
      (park_count * 3.0) + (museum_count * 1.5) + (family_restaurant_count * 2.5) AS family_friendliness_score,
      (park_count * 2.0) + (pet_service_count * 3.5) + (dog_friendly_place_count * 2.5) AS pet_paradise_score
    FROM
      insight_counts
  )
-- Final Step: Normalize scores to a 0-10 scale and rejoin to retrieve the location geometry.
SELECT
  raw_scores.id,
  raw_scores.name,
  apartments.location,
  raw_scores.family_friendliness_score,
  raw_scores.pet_paradise_score,
  -- Normalize Family Score using a MIN/MAX window function.
  ROUND(
    COALESCE(
      SAFE_DIVIDE(
        (raw_scores.family_friendliness_score - MIN(raw_scores.family_friendliness_score) OVER ()),
        (MAX(raw_scores.family_friendliness_score) OVER () - MIN(raw_scores.family_friendliness_score) OVER ())
      ) * 10,
      0
    ),
    2
  ) AS normalized_family_score,
  -- Normalize Pet Score using a MIN/MAX window function.
  ROUND(
    COALESCE(
      SAFE_DIVIDE(
        (raw_scores.pet_paradise_score - MIN(raw_scores.pet_paradise_score) OVER ()),
        (MAX(raw_scores.pet_paradise_score) OVER () - MIN(raw_scores.pet_paradise_score) OVER ())
      ) * 10,
      0
    ),
    2
  ) AS normalized_pet_score
FROM
  raw_scores
JOIN
  `your_project.your_dataset.apartment_listings` AS apartments
  ON raw_scores.id = apartments.id;

WITH
  -- CTE 1: Aggregate Place counts (Same as Step 2)
  insight_counts AS (
    SELECT WITH AGGREGATION_THRESHOLD
      apartments.id,
      apartments.name,
      apartments.description, -- Assuming your table has a description column
      COUNTIF(places.primary_type = 'park') AS park_count,
      COUNTIF(places.primary_type = 'museum') AS museum_count,
      COUNTIF(places.primary_type = 'restaurant' AND places.good_for_children = TRUE) AS family_restaurant_count
    FROM
      `your-project.your_dataset.apartment_listings` AS apartments
    LEFT JOIN
      `your-project.places_insights___us.places` AS places
      ON ST_DWITHIN(apartments.location, places.point, 800)
    GROUP BY
      apartments.id, apartments.name, apartments.description
  ),

  -- CTE 2: Construct the Prompt
  prepared_prompts AS (
    SELECT
      id,
      name,
      FORMAT("""
        You are an expert real estate analyst. Generate a 'Family-Friendliness Score' (0-10) for this location.

        Target User: Young family with a toddler, looking for a balance of activity and quiet.

        Location Data:
        - Name: %s
        - Description: %s
        - Parks nearby: %d
        - Museums nearby: %d
        - Family-friendly restaurants nearby: %d

        Scoring Rules:
        - High importance: Proximity to parks and high restaurant count.
        - Negative modifiers: Descriptions indicating excessive noise or nightlife focus.
        - Positive modifiers: Descriptions indicating quiet streets or backyards.
      """, name, description, park_count, museum_count, family_restaurant_count) AS prompt_text
    FROM insight_counts
    )

-- Final Step: Call AI.GENERATE
SELECT
  id,
  name,
  -- Access the structured fields returned by the model
  generated.family_friendliness_score,
  generated.reasoning
FROM
  prepared_prompts,
  AI.GENERATE(
    prompt_text,
    endpoint => 'gemini-flash-latest',
    output_schema => 'family_friendliness_score FLOAT64, reasoning STRING'
  ) AS generated;