Cleaning up code

clipft.py

@@ -1,5 +1,3 @@
-import asyncio
-import aiohttp
 import clip
 import logging
 import numpy as np
@@ -9,27 +7,6 @@
 from PIL import Image
 
 
-async def get(url, session):
-    headers = {
-        "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/42.0.2311.135 Safari/537.36 Edge/12.246"
-    }
-    try:
-        async with session.get(url=url, headers=headers) as response:
-            resp = await response.content.read()
-            return url, resp
-    except Exception as e:
-        logging.error(
-            "Unable to get url {} due to {}.".format(url, e.__class__)
-        )
-
-
-async def async_download_image(img_urls):
-    async with aiohttp.ClientSession() as session:
-        ret = await asyncio.gather(*[get(url, session) for url in img_urls])
-        ret = [x for x in ret if x is not None]
-        return [x[0] for x in ret], [x[1] for x in ret]
-
-
 class OutfitDataset(torch.utils.data.Dataset):
     def __init__(self, img_blobs, captions, device):
         _, preprocess_fn = clip.load("ViT-B/32", device=device)
@@ -101,7 +78,7 @@ def fine_tune_model(model, img_blobs, captions, batch_size=16, epochs=5):
     )
     best_loss = float("inf")
 
-    # Train model
+    # Train and validate model
     for epoch in range(epochs):
         # Train mode
         model.train()

dashboard.py

@@ -1,29 +1,31 @@
-import motion
 import streamlit as st
 
-import components
+from motion import get_store
+from transforms import SuggestIdea, Retrieval
+from schemas import QuerySchema, CatalogSchema, QuerySource
+from scrapers import scrape_everlane_sale
 
 
 @st.cache_resource
 def setup_database():
     # Create store and add triggers
-    store = motion.get_store("fashion", create=True, memory=True)
-    store.addNamespace("query", components.QuerySchema)
-    store.addNamespace("catalog", components.CatalogSchema)
+    store = get_store("fashion", create=True, memory=True)
+    store.addNamespace("query", QuerySchema)
+    store.addNamespace("catalog", CatalogSchema)
 
     store.addTrigger(
         name="suggest_idea",
         keys=["query.query"],
-        trigger=components.SuggestIdea,
+        trigger=SuggestIdea,
     )
     store.addTrigger(
         name="retrieval",
         keys=["catalog.img_blob", "query.text_suggestion", "query.feedback"],
-        trigger=components.Retrieval,
+        trigger=Retrieval,
     )
 
     # Add the catalog
-    components.scrape_everlane_sale(store, k=20)
+    scrape_everlane_sale(store, k=20)
 
     return store
 
@@ -37,7 +39,7 @@ def run_query(query):
         id=None,
         key_values={
             "query": query,
-            "src": components.QuerySource.ONLINE,
+            "src": QuerySource.ONLINE,
             "query_id": query_id,
         },
     )

motion/store.py

@@ -269,8 +269,8 @@ def executeTrigger(
             # Execute the transform lifecycle
             if trigger_fn.shouldFit(id, trigger_elem):
                 trigger_fn.fit(id, trigger_elem)
-            if trigger_fn.shouldTransform(id, trigger_elem):
-                trigger_fn.transform(id, trigger_elem)
+            if trigger_fn.shouldInfer(id, trigger_elem):
+                trigger_fn.infer(id, trigger_elem)
         logging.info(f"Finished running trigger {trigger_name}.")
 
     def set(

motion/transform.py

@@ -24,9 +24,9 @@ def fit(self, id, triggered_by: TriggerElement):
         pass
 
     @abstractmethod
-    def shouldTransform(self, id, triggered_by: TriggerElement):
+    def shouldInfer(self, id, triggered_by: TriggerElement):
         pass
 
     @abstractmethod
-    def transform(self, id, triggered_by: TriggerElement):
+    def infer(self, id, triggered_by: TriggerElement):
         pass

schemas.py

@@ -0,0 +1,33 @@
+import motion
+from typing import TypeVar
+
+
+class Retailer(motion.MEnum):
+    NORDSTROM = "Nordstrom"
+    REVOLVE = "Revolve"
+    BLOOMINGDALES = "Bloomingdales"
+    EVERLANE = "Everlane"
+
+
+class QuerySource(motion.MEnum):
+    OFFLINE = "Offline"
+    ONLINE = "Online"
+
+
+class QuerySchema(motion.Schema):
+    src: QuerySource
+    query_id: int
+    query: str
+    text_suggestion: str
+    img_id: int
+    img_score: float
+    feedback: bool
+
+
+class CatalogSchema(motion.Schema):
+    retailer: Retailer
+    img_url: str
+    img_blob: TypeVar("BLOB")
+    img_name: str
+    permalink: str
+    img_embedding: TypeVar("FLOAT[]")

scrapers.py

@@ -0,0 +1,87 @@
+import asyncio
+import logging
+import json
+import pandas as pd
+import requests
+
+from utils import async_download_image
+
+from bs4 import BeautifulSoup
+from schemas import Retailer
+
+
+def scrape_everlane_sale(store, k=20):
+    # Scrape the catalog and add the images to the store
+    urls = [
+        # "https://www.everlane.com/collections/womens-sale-2",
+        "https://www.everlane.com/collections/womens-all-tops",
+        "https://www.everlane.com/collections/womens-tees",
+        "https://www.everlane.com/collections/womens-sweaters",
+        "https://www.everlane.com/collections/womens-sweatshirts",
+        "https://www.everlane.com/collections/womens-bodysuits",
+        "https://www.everlane.com/collections/womens-jeans",
+        "https://www.everlane.com/collections/womens-bottoms",
+        "https://www.everlane.com/collections/womens-skirts-shorts",
+        "https://www.everlane.com/collections/womens-dresses",
+        "https://www.everlane.com/collections/womens-outerwear",
+        "https://www.everlane.com/collections/womens-underwear",
+        "https://www.everlane.com/collections/womens-perform",
+        "https://www.everlane.com/collections/swimwear",
+        "https://www.everlane.com/collections/womens-shoes",
+    ]
+    headers = {
+        "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/42.0.2311.135 Safari/537.36 Edge/12.246"
+    }
+    product_info = []
+    for url in urls:
+        r = requests.get(url=url, headers=headers)
+
+        soup = BeautifulSoup(r.content, "html5lib")
+
+        res = soup.find("script", attrs={"id": "__NEXT_DATA__"})
+        products = json.loads(res.contents[0])["props"]["pageProps"][
+            "fallbackData"
+        ]["products"]
+
+        for product in products:
+            img_url = product["albums"]["square"][0]["src"]
+            img_name = product["displayName"]
+            permalink = product["permalink"]
+            product_info.append(
+                {
+                    "img_url": img_url,
+                    "img_name": img_name,
+                    "permalink": "https://www.everlane.com/products/"
+                    + permalink,
+                }
+            )
+
+    # Delete duplicates
+    df = pd.DataFrame(product_info)
+    df = (
+        df.drop_duplicates(subset=["img_url"])
+        .sample(frac=1)
+        .reset_index(drop=True)
+    )
+    logging.info(f"Found {len(df)} unique products.")
+    df = df.head(k)
+
+    # Get blobs from the images
+    img_urls, contents = asyncio.run(
+        async_download_image(df["img_url"].values)
+    )
+    img_url_to_content = dict(zip(img_urls, contents))
+
+    for _, product_row in df.iterrows():
+        if product_row["img_url"] not in img_url_to_content:
+            continue
+
+        new_id = store.getNewId("catalog")
+        product = product_row.to_dict()
+        product.update(
+            {
+                "retailer": Retailer.EVERLANE,
+                "img_blob": img_url_to_content[product_row["img_url"]],
+            }
+        )
+        store.set("catalog", id=new_id, key_values=product)