Add personalised reco

app.py

@@ -0,0 +1,162 @@
+from typing import Dict, Text
+import numpy as np
+import tensorflow as tf
+import pandas as pd
+import pickle
+import numpy as np
+import tensorflow as tf
+import tensorflow_recommenders as tfrs
+import streamlit as st
+from html_information import html
+
+
+import pandas as pd
+import json
+
+def read_json(file_name):
+    with open(file_name) as json_file:
+        data = json.load(json_file)
+    return data
+
+uid_name_map = read_json('uid_name_map.json')
+uid_url_map = read_json('uid_url_map.json')
+
+
+st.set_page_config(page_title="My App", page_icon=":guardsman:", layout="wide", initial_sidebar_state="auto")
+
+@st.cache_resource
+def load_model(path):
+    loaded = tf.saved_model.load(path)
+    return loaded
+
+
+def inference(model, user_id):
+    scores, titles = model([user_id])
+    recs = titles[0, :15]
+    extracted_rec = []
+    for rec in recs:
+        extracted_rec.append(int(rec.numpy().decode('utf-8')))
+    return extracted_rec
+    
+    
+def read_pickle_file(file_path):
+    with open(file_path, 'rb') as f:
+        data = pickle.load(f)
+    return data
+
+def streamlit_carousel(header_name: str, rec_item_url: list,
+                       rec_item_name: list) -> None:
+    st.header(header_name)
+    mid_section = ""
+    for index, value in enumerate(rec_item_url):
+        mid_section += """<div class="item"><div id="image-container"><img src='""" + str(value) + """' /></div><p>""" + str(rec_item_name[index]) + """</p></div>"""
+
+    mid_html = html+mid_section + """</div></div></body>"""
+    st.markdown(mid_html, unsafe_allow_html=True)
+
+def recall_at_k(ground_truth, recommended, k):
+    """
+    Calculate Recall@k.
+
+    Parameters:
+    - ground_truth (list): List of ground truth product IDs.
+    - recommended (list): List of recommended product IDs.
+    - k (int): Number of recommendations to consider.
+
+    Returns:
+    - recall (float): Recall@k value.
+    """
+    # Take only the top-k recommended items
+    recommended_at_k = set(recommended[:k])
+    
+    # Count the number of relevant items in the ground truth
+    relevant_items = set(ground_truth)
+    
+    # Calculate the intersection (number of relevant items in top-k)
+    intersection = recommended_at_k.intersection(relevant_items)
+    
+    # Calculate Recall@k
+    recall = len(intersection) / len(relevant_items) if len(relevant_items) > 0 else 0.0
+    
+    return recall
+
+
+
+
+model_weights_name = 'gofynd_old_model.model'
+k = 15
+print("######## Running ########")
+print(f"model_weights_name: {model_weights_name}")
+print('########')
+print()
+loaded = load_model(model_weights_name)
+print("######### Model Loaded #########")
+
+
+# uid_name_map = read_pickle_file('new_uid_name_map.pkl')
+# uid_url_map = read_pickle_file('new_uid_url_map.pkl')
+# uid_url_map = 
+user_product_dict = read_pickle_file('user_product_dict.pkl')
+last_session_user_product_dict = read_pickle_file('final_sessions_fynd_pickle_filename.pkl')
+user_with_multiple_sessions = read_pickle_file('users_with_multiple_sessions_filename.pkl')
+initial_sessions_user_product_dict = read_pickle_file('initial_sessions_fynd_pickle_filename.pkl')
+# avg_recall = read_pickle_file('Personalised_two_tower_fynd_recall.pkl')
+# positive_recall = read_pickle_file("Personalised_two_twoer_fynd_positive_recall.pkl")
+# total_count = read_pickle_file("Personalised_two_twoer_fynd_total_count.pkl")
+# average_positive_recall = read_pickle_file("Personalised_two_twoer_fynd_average_positive_recall.pkl")
+user_id_list = user_with_multiple_sessions
+
+# # st.set_page_config(page_title="My App", page_icon=":guardsman:", layout="wide", initial_sidebar_state="auto")
+# st.header("Personalised Product Recommendations (Fynd)")
+# st.subheader("Training Metrics")
+# st.write(f"Average Recall@{k} on Test Set: {avg_recall}")
+# st.write(f"Total Users Count: {total_count}")
+# st.write(f"Users with Positive Recall@{k} on Test Set: {positive_recall}")
+# st.write(f"% Users with Positive Recall@{k} on Test Set: {average_positive_recall}")
+# col1, col2 = st.tabs(["Training & Test Loss", "Top 10 Test Accuracy"])
+# with col1:
+#     st.image('Personalised_two_tower_fynd_loss_graph.png')
+# with col2:
+#     st.image('Personalised_two_tower_fynd_top_10_accuracy_graph.png')
+    
+st.header("Personalised Product Recommendations")
+st.write("Model trained with Clickstream data of GoFynd.com")
+st.subheader("Choose a User")
+index = st.selectbox("User List", range(len(user_id_list)), format_func=lambda x: user_id_list[x])
+user_id = user_id_list[index]
+print(f"User ID: {user_id}")
+
+user_final_session = last_session_user_product_dict[user_id]
+final_session_product_list = []
+for all_session in user_final_session:
+    for session in all_session:
+        final_session_product_list.append(session['product_id'])
+
+rec_list = inference(loaded, str(user_id))
+print(f"Final Session Product List: {final_session_product_list}")
+print(f"Recommendation List: {rec_list}")
+recall_value = recall_at_k(final_session_product_list, rec_list, k)
+print(f"Recall@{k}: {recall_value}")
+st.write(f"Recommendation Score: {recall_value}")
+
+initial_sessions = initial_sessions_user_product_dict[user_id]
+
+tab1, tab2, tab3 = st.tabs(["Recommendations", "Test session data", "Train session data"])
+with tab1:
+    rec_list_name = [uid_name_map[product_id] for product_id in rec_list]
+    rec_list_url = [uid_url_map[product_id] for product_id in rec_list]
+    streamlit_carousel("Top 15 Personalised Product Recommendation", rec_list_url, rec_list_name)
+with tab2:
+    product_name_list = [uid_name_map[product_id] for product_id in final_session_product_list]
+    product_url_list = [uid_url_map[product_id] for product_id in final_session_product_list]
+    streamlit_carousel("User's Test Last Session Viewed Products", product_url_list, product_name_list)
+with tab3:
+    i=1
+    for session in initial_sessions:
+        temp_product_list = []
+        for row in session:
+            temp_product_list.append(row['product_id'])
+        product_name_list = [uid_name_map[product_id] for product_id in temp_product_list]
+        product_url_list = [uid_url_map[product_id] for product_id in temp_product_list]
+        streamlit_carousel("Session "+str(i), product_url_list, product_name_list)
+        i+=1

gofynd_old_model.model/fingerprint.pb

@@ -0,0 +1 @@
+����ニ�,ꓼ�ߍ�����‡�� م������
(ȟ�궷�i2

grouping_data.py

@@ -0,0 +1,24 @@
+import pandas as pd
+
+df = pd.read_csv('new_events_data.csv')
+grouped_data = df.groupby('user_id')['product_id'].agg(list).reset_index()
+
+user_product_dict = dict(zip(grouped_data['user_id'], grouped_data['product_id']))
+
+for key, value in user_product_dict.items():
+    print(key, value)
+    break
+
+user_id_list = list(user_product_dict.keys())
+print(len(user_id_list))
+print(len(user_product_dict))
+
+import pickle
+
+def save_obj(obj, name):
+    with open(name, 'wb') as f:
+        pickle.dump(obj, f)
+        
+
+save_obj(user_product_dict, 'user_product_dict.pkl')
+save_obj(user_id_list, 'user_id_list.pkl')

html_information.py

@@ -0,0 +1,67 @@
+html = """
+<style>
+ body{
+  font-family: sans-serif;
+}
+h1{
+  font-weight: 100;
+}
+.container{
+  width: 100%;
+  display:block;
+  overflow:hidden;
+}
+.carousel{
+  display:block;
+  width: 100%;
+  height: 320px;
+  background: white;
+  overflow-x: scroll;
+  padding: 10px;
+  margin: 0;
+  white-space: nowrap; 
+  border-top: 2px solid rgba(0, 0, 0, 0.1);
+  border-bottom: 2px solid rgba(0, 0, 0, 0.1);
+}
+.item {
+    display: inline-block;
+    overflow: hidden;
+    width: 250px;
+    margin: 0 10px;
+    height: calc(100%);
+    background: rgba(0, 0, 0, 0.05) no-repeat center center;
+    background-size: cover;
+    position:relative;
+    border-radius: 20px;
+    box-shadow: 0 0 10px #dfdfdf;
+}
+.item p {
+    padding: 20px;
+    word-break: break-all;
+    white-space: break-spaces;
+    overflow: hidden;
+    display: -webkit-box;
+    -webkit-line-clamp: 4;
+    -webkit-box-orient: vertical;
+    margin: 0;
+    align: center;
+}
+#image-container{
+  width: 100%;
+  height: 60%;
+  text-align:center;
+  font-size: 9em;
+  color: white;
+  overflow: hidden;
+}
+#image-container img{
+    width: 100%;
+    height: 100%;
+    object-fit: contain;
+}
+</style>
+</head>
+<body>
+<div class="container">
+  <div class="carousel">
+"""

requirements.txt

@@ -0,0 +1,9 @@
+keras==2.15.0
+tensorboard==2.15.1
+tensorboard-data-server==0.7.2
+tensorflow==2.15.0
+tensorflow-estimator==2.15.0
+tensorflow-io-gcs-filesystem==0.34.0
+tensorflow-recommenders==0.7.3
+gcsfs
+streamlit

url_mapper.py

@@ -0,0 +1,53 @@
+import pandas as pd
+import json
+
+def read_json(file_name):
+    with open(file_name) as json_file:
+        data = json.load(json_file)
+    return data
+
+catalog_data = read_json('catalog.json')
+
+inst = catalog_data[0]
+print("Catalog data")
+print(inst)
+
+for key, value in inst.items():
+    print("Key: ", key)
+    print("Value: ", value)
+    print()
+
+
+
+
+for key, value in inst.items():
+    if key == "medias":
+        print(value[0]['url'])
+    if key == 'name':
+        print(value)
+        
+uid_name_map = {}
+uid_url_map = {}
+
+for inst in catalog_data:
+    for key, value in inst.items():
+        if key == "medias":
+            uid_url_map[inst['uid']] = value[0]['url']
+        if key == 'name':
+            uid_name_map[inst['uid']] = value
+
+print(len(uid_name_map))
+print(len(uid_url_map))
+with open('uid_name_map.json', 'w') as json_file:
+    json.dump(uid_name_map, json_file, indent=4)
+with open('uid_url_map.json', 'w') as json_file:
+    json.dump(uid_url_map, json_file, indent=4)
+
+# import pickle
+
+# def save_obj(obj, name):
+#     with open(name, 'wb') as f:
+#         pickle.dump(obj, f)
+        
+# save_obj(uid_name_map, 'new_uid_name_map.pkl')
+# save_obj(uid_url_map, 'new_uid_url_map.pkl')