connect to pw-recsys to get rankings for user
Browse files
app.py
CHANGED
|
@@ -4,6 +4,8 @@ from pyvis.network import Network
|
|
| 4 |
import pickle
|
| 5 |
import math
|
| 6 |
import random
|
|
|
|
|
|
|
| 7 |
|
| 8 |
# Dictionary to map brands to their respective pickle files
|
| 9 |
BRAND_GRAPHS = {
|
|
@@ -13,18 +15,18 @@ BRAND_GRAPHS = {
|
|
| 13 |
'guitareo': 'guitareo_pop_items_labels.pkl'
|
| 14 |
}
|
| 15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
@st.cache_resource
|
| 17 |
def load_graph(brand):
|
| 18 |
-
"""
|
| 19 |
-
Load the graph for the selected brand.
|
| 20 |
-
"""
|
| 21 |
with open(BRAND_GRAPHS[brand], 'rb') as f:
|
| 22 |
return pickle.load(f)
|
| 23 |
|
| 24 |
def filter_graph(graph, node_threshold=10, edge_threshold=5):
|
| 25 |
-
"""
|
| 26 |
-
Filters the graph to include only popular nodes and edges.
|
| 27 |
-
"""
|
| 28 |
popular_nodes = [
|
| 29 |
node for node in graph.nodes
|
| 30 |
if graph.degree(node) >= node_threshold
|
|
@@ -38,7 +40,49 @@ def filter_graph(graph, node_threshold=10, edge_threshold=5):
|
|
| 38 |
|
| 39 |
return filtered_graph
|
| 40 |
|
| 41 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
net = Network(notebook=False, width="100%", height="600px", directed=True)
|
| 43 |
net.set_options("""
|
| 44 |
var options = {
|
|
@@ -55,18 +99,27 @@ def dynamic_visualize_graph(graph, start_node, layers=3, top_k=5, show_titles=Fa
|
|
| 55 |
added_edges = set()
|
| 56 |
current_nodes = [int(start_node)]
|
| 57 |
|
|
|
|
|
|
|
| 58 |
# Add the starting node, color it red, and include a tooltip
|
| 59 |
start_title = graph.nodes[int(start_node)].get('title', 'No title available')
|
| 60 |
start_in_degree = graph.in_degree(int(start_node))
|
| 61 |
start_out_degree = graph.out_degree(int(start_node))
|
| 62 |
start_node_size = (start_in_degree + start_out_degree) * 0.15
|
| 63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 64 |
net.add_node(
|
| 65 |
int(start_node),
|
| 66 |
label=label,
|
| 67 |
-
color="darkblue",
|
| 68 |
-
title=f"{start_title} In-degree: {start_in_degree}, Out-degree: {start_out_degree}",
|
| 69 |
-
size=start_node_size
|
|
|
|
|
|
|
| 70 |
)
|
| 71 |
visited_nodes.add(int(start_node))
|
| 72 |
|
|
@@ -85,15 +138,24 @@ def dynamic_visualize_graph(graph, start_node, layers=3, top_k=5, show_titles=Fa
|
|
| 85 |
neighbor_in_degree = graph.in_degree(neighbor)
|
| 86 |
neighbor_out_degree = graph.out_degree(neighbor)
|
| 87 |
neighbor_size = (neighbor_in_degree + neighbor_out_degree) * 0.15
|
|
|
|
|
|
|
| 88 |
node_color = 'red' if neighbor_in_degree > neighbor_out_degree * 1.5 else \
|
| 89 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 90 |
label = str(neighbor) if not show_titles else f"{str(neighbor)}: {neighbor_title[:15]}..."
|
| 91 |
net.add_node(
|
| 92 |
neighbor,
|
| 93 |
label=label,
|
| 94 |
-
title=f"{neighbor_title} In-degree: {neighbor_in_degree}, Out-degree: {neighbor_out_degree}",
|
| 95 |
size=neighbor_size,
|
| 96 |
-
color=node_color
|
|
|
|
|
|
|
| 97 |
)
|
| 98 |
edge = (node, neighbor)
|
| 99 |
if edge not in added_edges:
|
|
@@ -109,7 +171,7 @@ def dynamic_visualize_graph(graph, start_node, layers=3, top_k=5, show_titles=Fa
|
|
| 109 |
st.components.v1.html(html_content, height=600, scrolling=False)
|
| 110 |
|
| 111 |
|
| 112 |
-
st.title("Popular Path Expansion")
|
| 113 |
|
| 114 |
# Brand Selection
|
| 115 |
selected_brand = st.selectbox("Select a brand:", options=list(BRAND_GRAPHS.keys()))
|
|
@@ -136,6 +198,10 @@ start_node = st.number_input(
|
|
| 136 |
step=1
|
| 137 |
)
|
| 138 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 139 |
# Toggle for showing content titles
|
| 140 |
show_titles = st.checkbox("Show content titles", value=False)
|
| 141 |
|
|
@@ -144,11 +210,19 @@ node_degree_threshold = 1
|
|
| 144 |
edge_weight_threshold = 1
|
| 145 |
G_filtered = filter_graph(G, node_threshold=node_degree_threshold, edge_threshold=edge_weight_threshold)
|
| 146 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 147 |
layers = st.slider("Depth to explore:", 1, 6, value=3)
|
| 148 |
top_k = st.slider("Branching factor (per node):", 1, 6, value=3)
|
| 149 |
|
| 150 |
if st.button("Expand Graph"):
|
| 151 |
if start_node in G_filtered:
|
| 152 |
-
dynamic_visualize_graph(G_filtered, start_node, layers=layers, top_k=top_k, show_titles=show_titles)
|
| 153 |
else:
|
| 154 |
st.error("The starting node is not in the graph!")
|
|
|
|
| 4 |
import pickle
|
| 5 |
import math
|
| 6 |
import random
|
| 7 |
+
import requests
|
| 8 |
+
import os
|
| 9 |
|
| 10 |
# Dictionary to map brands to their respective pickle files
|
| 11 |
BRAND_GRAPHS = {
|
|
|
|
| 15 |
'guitareo': 'guitareo_pop_items_labels.pkl'
|
| 16 |
}
|
| 17 |
|
| 18 |
+
# API Authorization Token
|
| 19 |
+
AUTH_TOKEN = os.getenv('HF_TOKEN')
|
| 20 |
+
|
| 21 |
+
API_URL = "https://MusoraProductDepartment-PWGenerator.hf.space/rank_items/"
|
| 22 |
+
|
| 23 |
+
|
| 24 |
@st.cache_resource
|
| 25 |
def load_graph(brand):
|
|
|
|
|
|
|
|
|
|
| 26 |
with open(BRAND_GRAPHS[brand], 'rb') as f:
|
| 27 |
return pickle.load(f)
|
| 28 |
|
| 29 |
def filter_graph(graph, node_threshold=10, edge_threshold=5):
|
|
|
|
|
|
|
|
|
|
| 30 |
popular_nodes = [
|
| 31 |
node for node in graph.nodes
|
| 32 |
if graph.degree(node) >= node_threshold
|
|
|
|
| 40 |
|
| 41 |
return filtered_graph
|
| 42 |
|
| 43 |
+
def get_rankings_from_api(brand, user_id, content_ids):
|
| 44 |
+
"""
|
| 45 |
+
Call the rank_items API to fetch rankings for the given user and content IDs.
|
| 46 |
+
"""
|
| 47 |
+
try:
|
| 48 |
+
payload = {
|
| 49 |
+
"brand": brand.upper(),
|
| 50 |
+
"user_id": user_id,
|
| 51 |
+
"content_ids": content_ids
|
| 52 |
+
}
|
| 53 |
+
headers = {
|
| 54 |
+
"Authorization": f"Bearer {AUTH_TOKEN}",
|
| 55 |
+
"accept": "application/json",
|
| 56 |
+
"Content-Type": "application/json"
|
| 57 |
+
}
|
| 58 |
+
response = requests.post(API_URL, json=payload, headers=headers)
|
| 59 |
+
response.raise_for_status()
|
| 60 |
+
rankings = response.json()
|
| 61 |
+
return rankings
|
| 62 |
+
except Exception as e:
|
| 63 |
+
st.error(f"Error calling rank_items API: {e}")
|
| 64 |
+
return {}
|
| 65 |
+
|
| 66 |
+
|
| 67 |
+
def rank_to_color(rank, max_rank):
|
| 68 |
+
"""
|
| 69 |
+
Map a rank to a grayscale color, where dark gray indicates high relevance (low rank),
|
| 70 |
+
and light gray indicates low relevance (high rank).
|
| 71 |
+
|
| 72 |
+
Parameters:
|
| 73 |
+
rank (int): The rank of the item.
|
| 74 |
+
max_rank (int): The maximum rank (for normalizing the gradient).
|
| 75 |
+
|
| 76 |
+
Returns:
|
| 77 |
+
str: Hex color code for the grayscale shade.
|
| 78 |
+
"""
|
| 79 |
+
if rank > max_rank: # Handle items without ranking
|
| 80 |
+
return "#E8E8E8" # Very light gray for unranked items
|
| 81 |
+
intensity = int(55 + (rank / max_rank) * 200) # Scale intensity (darker for lower ranks)
|
| 82 |
+
return f"rgb({intensity}, {intensity}, {intensity})" # Grayscale
|
| 83 |
+
|
| 84 |
+
|
| 85 |
+
def dynamic_visualize_graph(graph, start_node, layers=3, top_k=5, show_titles=False, rankings=None):
|
| 86 |
net = Network(notebook=False, width="100%", height="600px", directed=True)
|
| 87 |
net.set_options("""
|
| 88 |
var options = {
|
|
|
|
| 99 |
added_edges = set()
|
| 100 |
current_nodes = [int(start_node)]
|
| 101 |
|
| 102 |
+
max_rank = len(rankings) if rankings else 0
|
| 103 |
+
|
| 104 |
# Add the starting node, color it red, and include a tooltip
|
| 105 |
start_title = graph.nodes[int(start_node)].get('title', 'No title available')
|
| 106 |
start_in_degree = graph.in_degree(int(start_node))
|
| 107 |
start_out_degree = graph.out_degree(int(start_node))
|
| 108 |
start_node_size = (start_in_degree + start_out_degree) * 0.15
|
| 109 |
+
start_rank = rankings.index(int(start_node)) if rankings and int(start_node) in rankings else max_rank + 1
|
| 110 |
+
if rankings:
|
| 111 |
+
start_border_color = rank_to_color(start_rank, max_rank)
|
| 112 |
+
else:
|
| 113 |
+
start_border_color = 'darkblue'
|
| 114 |
+
label = str(start_node) if not show_titles else f"{str(start_node)}: {start_title[:15]}..."
|
| 115 |
net.add_node(
|
| 116 |
int(start_node),
|
| 117 |
label=label,
|
| 118 |
+
color={"background": "darkblue", "border": start_border_color},
|
| 119 |
+
title=f"{start_title}, In-degree: {start_in_degree}, Out-degree: {start_out_degree}, Rank: {start_rank}",
|
| 120 |
+
size=start_node_size,
|
| 121 |
+
borderWidth=3,
|
| 122 |
+
borderWidthSelected=6
|
| 123 |
)
|
| 124 |
visited_nodes.add(int(start_node))
|
| 125 |
|
|
|
|
| 138 |
neighbor_in_degree = graph.in_degree(neighbor)
|
| 139 |
neighbor_out_degree = graph.out_degree(neighbor)
|
| 140 |
neighbor_size = (neighbor_in_degree + neighbor_out_degree) * 0.15
|
| 141 |
+
neighbor_rank = rankings.index(neighbor) if rankings and neighbor in rankings else max_rank + 1
|
| 142 |
+
|
| 143 |
node_color = 'red' if neighbor_in_degree > neighbor_out_degree * 1.5 else \
|
| 144 |
+
'green' if neighbor_out_degree > neighbor_in_degree * 1.5 else 'lightblue'
|
| 145 |
+
if rankings:
|
| 146 |
+
neighbor_border_color = rank_to_color(neighbor_rank, max_rank)
|
| 147 |
+
else:
|
| 148 |
+
neighbor_border_color = node_color
|
| 149 |
+
|
| 150 |
label = str(neighbor) if not show_titles else f"{str(neighbor)}: {neighbor_title[:15]}..."
|
| 151 |
net.add_node(
|
| 152 |
neighbor,
|
| 153 |
label=label,
|
| 154 |
+
title=f"{neighbor_title}, In-degree: {neighbor_in_degree}, Out-degree: {neighbor_out_degree}, Rank: {neighbor_rank}",
|
| 155 |
size=neighbor_size,
|
| 156 |
+
color={"background": node_color, "border": neighbor_border_color},
|
| 157 |
+
borderWidth=3,
|
| 158 |
+
borderWidthSelected=6
|
| 159 |
)
|
| 160 |
edge = (node, neighbor)
|
| 161 |
if edge not in added_edges:
|
|
|
|
| 171 |
st.components.v1.html(html_content, height=600, scrolling=False)
|
| 172 |
|
| 173 |
|
| 174 |
+
st.title("Popular Path Expansion + Personalization")
|
| 175 |
|
| 176 |
# Brand Selection
|
| 177 |
selected_brand = st.selectbox("Select a brand:", options=list(BRAND_GRAPHS.keys()))
|
|
|
|
| 198 |
step=1
|
| 199 |
)
|
| 200 |
|
| 201 |
+
# Input: Student ID
|
| 202 |
+
student_id = st.text_input("Enter a student ID (optional):", value="")
|
| 203 |
+
|
| 204 |
+
|
| 205 |
# Toggle for showing content titles
|
| 206 |
show_titles = st.checkbox("Show content titles", value=False)
|
| 207 |
|
|
|
|
| 210 |
edge_weight_threshold = 1
|
| 211 |
G_filtered = filter_graph(G, node_threshold=node_degree_threshold, edge_threshold=edge_weight_threshold)
|
| 212 |
|
| 213 |
+
# Fetch rankings if student ID is provided
|
| 214 |
+
rankings = {}
|
| 215 |
+
if student_id:
|
| 216 |
+
content_ids = list(G_filtered.nodes)
|
| 217 |
+
rankings = get_rankings_from_api(selected_brand, int(student_id), content_ids)
|
| 218 |
+
if rankings:
|
| 219 |
+
rankings = rankings['ranked_content_ids']
|
| 220 |
+
#print(rankings)
|
| 221 |
layers = st.slider("Depth to explore:", 1, 6, value=3)
|
| 222 |
top_k = st.slider("Branching factor (per node):", 1, 6, value=3)
|
| 223 |
|
| 224 |
if st.button("Expand Graph"):
|
| 225 |
if start_node in G_filtered:
|
| 226 |
+
dynamic_visualize_graph(G_filtered, start_node, layers=layers, top_k=top_k, show_titles=show_titles, rankings=rankings)
|
| 227 |
else:
|
| 228 |
st.error("The starting node is not in the graph!")
|