first commit

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.data-00000-of-00001 filter=lfs diff=lfs merge=lfs -text
+*.snapshot filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1 @@
+__pycache__

README.md

@@ -1,8 +1,8 @@
 ---
-title: Demo Recs Engine
-emoji: 👀
-colorFrom: yellow
-colorTo: gray
+title: Recommender Engine
+emoji: 🤗
+colorFrom: pink
+colorTo: green
 sdk: streamlit
 sdk_version: 1.38.0
 app_file: app.py

app.py

@@ -0,0 +1,65 @@
+import streamlit as st
+import pandas as pd
+from datetime import datetime
+from inference import Inference
+
+
+class QueryInputForm:
+    def __init__(self):
+        # Title of the Streamlit form
+        st.title("Query Input Form")
+
+        # Predefined options for channel and device type
+        self.channel_options = [
+            'Paid Social', 'Paid Search - Brand', 'Organic', 'Email - Transactional',
+            'Affiliate', 'Paid Search', 'Direct', 'Referral', 'Email - Marketing',
+            'Paid Search - Brand Reactivation', 'SMS - Marketing', 'Email - Trigger',
+            'Referral - Whitelabel', 'Referral - Merchant', 'Social', 'SMS - Trigger',
+        ]
+        
+        self.device_type_options = [
+            'Mobile', 'Desktop', 'Phablet', 'Tablet', '', 'TV',
+            'Portable Media Player', 'Wearable',
+        ]
+        
+        # Default values for the form
+        self.default_user_id = "b7485193f4e7f5b8ac3c94f71f4456a9"
+        self.default_query_text = "pizza"
+
+        # Initialize the recommender engine
+        self.recommender_engine = Inference()
+
+    def display_form(self):
+        # Input fields for user ID, channel, device type, and query text
+        self.user_id = st.text_input("User ID", value=self.default_user_id)
+        self.channel = st.selectbox("Channel", options=self.channel_options)
+        self.device_type = st.selectbox("Device Type", options=self.device_type_options)
+        self.query_text = st.text_input("Query Text", value=self.default_query_text)
+
+        # Submit button
+        if st.button("Submit"):
+            self.submit()
+
+    def submit(self):
+        # Pass the query information to the recommender engine
+        raw_query = {
+            'user_id': self.user_id,
+            'channel': self.channel,
+            'device_type': self.device_type,
+            'query_text': self.query_text,
+            'time': datetime.now().strftime("%Y-%m-%d %H:%M:%S.%f"),
+        }
+
+        # Get recommendations
+        self.recommender_engine.get_recommendations(raw_query)
+        self.display_recommendations()
+
+    def display_recommendations(self):
+        # Output the recommendations from the inference engine
+        st.write("Top recommendations:")
+        st.dataframe(self.recommender_engine.recommendations, hide_index=True)
+
+
+if __name__ == "__main__":
+    form = QueryInputForm()
+    form.display_form()

inference.py

@@ -0,0 +1,87 @@
+import tensorflow as tf
+import os
+
+
+class Inference():
+    def __init__(self):
+   
+        index_path = os.path.join("recommender_model", "index")
+        model_path = os.path.join("recommender_model", "model")
+        self.index = tf.keras.models.load_model(index_path)
+        self.model = tf.keras.models.load_model(model_path)
+        
+        products_path = os.path.join("products")
+        self.products = tf.data.Dataset.load(products_path)
+        
+        self.model_product_features = [
+            'product_id', 'category_name', 'merchant_name', 'merchant_city', 'merchant_state', 'merchant_region', 
+            'free_shipping', 'is_sold_out', 'editor_pick', 'on_sale', 'product_name', 'price_in_cents', 'reviews'
+        ]
+
+    def get_recommendations(self, raw_query: dict):
+        self.query_input = {
+            'user_id': tf.convert_to_tensor(raw_query['user_id'], dtype=tf.string),
+            'channel': tf.convert_to_tensor(raw_query['channel'], dtype=tf.string),
+            'device_type': tf.convert_to_tensor(raw_query['device_type'], dtype=tf.string),
+            'query_text': tf.convert_to_tensor(raw_query['query_text'], dtype=tf.string),
+            'time': tf.convert_to_tensor(raw_query['time'], dtype=tf.string),
+        }
+
+        # Get recommendations. Note that I am expanding the dimension to match the batch size expected by the model
+        _, self.top_rec = self.index({k: [v] for k, v in self.query_input.items()}) 
+    
+        # Filter by product id
+        filtered_recs = self.products.filter(self.filter_by_id)
+        # Add query input
+        query_added_recs = filtered_recs.map(lambda x: {**self.query_input, **x})
+        # Get score
+        score_added_recs = query_added_recs.batch(8).map(self.get_score).unbatch()
+
+        # Drop unwanted columns
+        recs = score_added_recs.map(self.desired_output)
+
+        # Order by score
+        ordered_recs = self.order_by_score(recs)
+
+        # Decode values
+        self.recommendations = list(map(self.decode_values, ordered_recs))
+
+    def filter_by_id(self, item):
+        return tf.reduce_any(tf.equal(item['product_id'], self.top_rec[0]))
+
+    def get_score(self, item):
+        input_data = {k: v for k, v in item.items() if k in self.model_product_features + list(self.query_input.keys())}
+        _, _, score = self.model(input_data)
+        item['score'] = score
+        return item
+    
+    def desired_output(self, item):
+        return {
+            'product_name': item['product_name'],
+            'category_name': item['category_name'],
+            'price_in_cents': item['price_in_cents'],
+            'reviews': item['reviews'],
+            'merchant_name': item['merchant_name'],
+            'merchant_city': item['merchant_city'],
+            'merchant_state': item['merchant_state'],
+            'merchant_region': item['merchant_region'],
+            'free_shipping': item['free_shipping'],
+            'is_sold_out': item['is_sold_out'],
+            'editor_pick': item['editor_pick'],
+            'on_sale': item['on_sale'],
+            'score': item['score']
+        }
+
+    def order_by_score(self, recs):
+        rec_list = list(recs.as_numpy_iterator())
+
+        # Descending order by score
+        return sorted(rec_list, key=lambda x: x['score'], reverse=True)
+
+    def decode_values(self, item):
+        for key, value in item.items():
+            if isinstance(value, bytes):
+                item[key] = value.decode('utf-8')
+            if key == 'score':
+                item[key] = value[0]
+        return item

products/4470620803936518297/00000000.shard/00000000.snapshot

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6b7c352940d95107a6bfc9035e75d6797ad9f18449205e1f718ec3757835065
+size 5133290

products/7776351335182402423/00000000.shard/00000000.snapshot

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6b7c352940d95107a6bfc9035e75d6797ad9f18449205e1f718ec3757835065
+size 5133290

products/dataset_spec.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6e96cdf4683d9d39dfc6f9479faacaf67e96d9e82abf87731908391a2b1b55d
+size 411

products/snapshot.metadata

@@ -0,0 +1 @@
+4470620803936518297�ޘ��� *				0�H�>

recommender_model/index/fingerprint.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5fa78657cc3dc078663289fc7e2e2269269e30855679fa8dd8fe6ecc1cc784e3
+size 56

recommender_model/index/keras_metadata.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:feef2be744045e53d44c33c63dd223c3d14a1fe0d57db1044a5d8c3506bf8fb4
+size 43956

recommender_model/index/saved_model.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54aa27f06f6c7bc92493e16f14fdc4c0d5a4418e68f5a7be3900a986167cd345
+size 1340342

recommender_model/index/variables/variables.data-00000-of-00001

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8c214a6184250fe8cd49e17760e466ae3aafd7d952cd92338d754183dca2fea
+size 21942235

recommender_model/model/fingerprint.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73dd5eb15d19b0f7ba30b2974616716ad220ab358f4f5a0ccb63f4a1abb91441
+size 56

recommender_model/model/keras_metadata.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f5a6e2a033359beeaa8ea2aaa03b19a37affcc7b0ee9be7383acf6d487bec32
+size 114938

recommender_model/model/saved_model.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ac62f41927afd25da1c115920addf4834d98db64c5a11dbfcb48307e3b5edd5
+size 2888902

recommender_model/model/variables/variables.data-00000-of-00001

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:469452fc885fbaa976c2fc4867844c1b77258c1f4265a75699fc2a29316091fe
+size 37821029

requirements.txt

@@ -0,0 +1,2 @@
+pandas
+tensorflow==2.15.1