Merge branch 'main' of https://huggingface.co/spaces/thinkcol/restaurants into main

app.py

@@ -68,7 +68,7 @@ async def on_message(message: str, message_id: str):
 
         await chain.text(str(list(results)))  # TODO maybe json format would be better?
 
-        restaurants = "\n".join(f"- ID: {r.id} | Description: {r.text}" for r in results)
+        restaurants = "\n".join(f"- ID: {r.id} | {r.text}" for r in results)
 
         final_choices_msg = await chain.llm(
             """
@@ -149,7 +149,7 @@ async def on_message(message: str, message_id: str):
             msg.elements = [
                 # note: image always displays above text
                 cl.Image(name=restaurant.name, url=restaurant.image_url, display='inline', size='small'),
-                cl.Text(name=restaurant.name, content=restaurant.text, display='inline'),
+                cl.Text(name=restaurant.name, content=restaurant.intro, display='inline'),
                 cl.Text(name="Example Dishes:", content=dishes_as_string, display='inline'),
                 cl.Text(name="Category:", content=categories_as_string, display='inline'),
                 cl.Text(name="Estimated Average Price (HKD):", content=restaurant.price, display="inline"),

data.py

@@ -45,12 +45,14 @@ class RestaurantDescription(OpenaiEmbeddingDoc):
     id: str = ''  # a number string
     name: str
     name_alt: str | None
+    intro: str
     categories: list[str]
     dishes: list[str]
     rating: float  # 0-1
     price: int  # HKD
     info_url: str
     image_url: str
+    location: list[str]
 
 
 class Category(OpenaiEmbeddingDoc):

data/categories.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8b67b72a7c15a455a0ef466c7cecd32295ecde6bf439bc78eac295f52fb6af35
-size 1135560
+oid sha256:4dddf959c29d7af260e7eeacd5a6dbd1c249b54fc492b21c6489dcf68f970402
+size 1134750

data/dishes.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6a86200c149121da1192572f3c7d4e395d38c91ac324b999d3aa9caecfd92b1
-size 7290545
+oid sha256:5e8436a53e09bff69275cd04b47a46c55a1e4e6e4f8ae1f856026e02060bffd9
+size 7290051

data/restaurants.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a88e0c7ccbb48546152e834d24933fe12e2785ab367cfce1cee8144e8ddfb51a
-size 2091841
+oid sha256:9d379bf61329a9e615a2cdfe082e99df5fe63f3278876e5addcf18e6e059c4ec
+size 1863120

scripts/create_embeddings.py

@@ -1,4 +1,5 @@
 import csv
+import json
 from ast import literal_eval
 from pathlib import Path
 from typing import TypeVar
@@ -35,6 +36,35 @@ def add_to_all(restaurant: RestaurantDescription, keys: list[str], mapping: dict
         v.restaurants.append(restaurant.id)
 
 
+def load_districts():
+    with Path('data/district_boundary.json').open('r', encoding='utf-8') as f:
+        districts = json.load(f)
+    from matplotlib.path import Path as Polyline
+    output = []
+    for d in districts['features']:
+        district = {
+            "name": d['properties']['District'],
+            "polygon": Polyline(d['geometry']['coordinates'][0])
+        }
+        output.append(district)
+
+    return output
+
+
+DISTRICTS = load_districts()
+
+
+def get_district_name(lat: str, lon: str):
+    lat = float(lat)
+    lon = float(lon)
+
+    matches = []
+    for district in DISTRICTS:
+        if district['polygon'].contains_point((lon, lat)):
+            matches.append(district['name'])
+    return matches
+
+
 restaurants, dish_list, category_list = None, None, None
 
 
@@ -45,6 +75,7 @@ def main():
     csv_file = Path('restaurants.csv')
 
     restaurants = DocList[RestaurantDescription]()
+    restaurant_names = set()
     dishes = {}
     categories = {}
 
@@ -58,30 +89,51 @@ def main():
                 name = row['name_lang1']
                 name_alt = None
 
+            # for this demo, don't add multiple locations of the same restaurant chain
+            if name in restaurant_names:
+                continue
+
             ds = literal_eval(row['dishes'])
             ds = [normalize_dish(d) for d in ds]
+            ds = list(set(ds))  # unique
             cs = literal_eval(row['categories'])
 
+            location = get_district_name(row['map_latitude'], row['map_longitude'])
+
+            text = f"""\
+Name: {name}
+Intro: {row['intro']}
+Categories: {", ".join(cs)}
+Dishes: {", ".join(ds)}
+Location: {", ".join(location)}\
+"""
+
             r = RestaurantDescription(
                 embedding=None,  # batch create all embeddings later
+                text=text,
                 id=row['id'],
                 name=name,
                 name_alt=name_alt,
-                text=row['intro'],
+                intro=row['intro'],
                 price=int(row['price']),
                 rating=calculate_rating(row['score_cry'], row['score_o_k'], row['score_smile']),
                 categories=cs,
                 dishes=ds,
                 info_url=row['poi_url'],
                 image_url=row['door_photos'],
+                location=location,
             )
 
             restaurants.append(r)
+            restaurant_names.add(name)
             add_to_all(r, ds, dishes, Dish)
             add_to_all(r, cs, categories, Category)
     dish_list = DocList[Dish](dishes.values())
     category_list = DocList[Category](categories.values())
 
+    # import IPython
+    # IPython.embed()
+
     embedding_settings = AzureOpenaiEmbeddings.load_from_env()
 
     RestaurantDescription.create_embeddings(restaurants, **embedding_settings.to_settings_dict())