File size: 6,601 Bytes
38ae75d
 
 
 
 
f91b841
38ae75d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c6233c8
38ae75d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
import gradio as gr
import torch
import numpy as np
import pandas as pd
from datetime import datetime, timedelta
from models import SASRec
from data_prepare import SASRecDataset, SASRecDataModule, prepare_data
from utils import load_item_properties, load_category_tree, get_popular_items

# --- Global variables to hold loaded artifacts ---
# This prevents reloading the model and data on every prediction.
MODEL = None
DATAMODULE = None
ITEM_CATEGORY_MAP = None
CATEGORY_PARENT_MAP = None
POPULAR_ITEMS = None

# --- Data Loading and Preparation Functions ---

def load_artifacts():
    """
    Loads all necessary artifacts (model, data, mappings) into global variables.
    This function is called only once when the app starts.
    """
    global MODEL, DATAMODULE, ITEM_CATEGORY_MAP, CATEGORY_PARENT_MAP, POPULAR_ITEMS
    
    print("--- Loading all artifacts for the Gradio app ---")
    
    # HF-FRIENDLY: Path is relative, assuming the checkpoint is in the root of the Space repo.
    CHECKPOINT_PATH = "checkpoints/sasrec-epoch=06-val_hitrate@10=0.3629.ckpt"
    DATA_FOLDER = "data/"

    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    print(f"Using device: {device}")

    print(f"Loading model from checkpoint: {CHECKPOINT_PATH}...")
    MODEL = SASRec.load_from_checkpoint(CHECKPOINT_PATH)
    MODEL.to(device)
    MODEL.eval()

    print("Preparing data...")
    train_set, validation_set, test_set = prepare_data(data_folder=DATA_FOLDER)
    
    DATAMODULE = SASRecDataModule(train_set, validation_set, test_set)
    DATAMODULE.setup()
    
    print("Loading item and category maps...")
    ITEM_CATEGORY_MAP = load_item_properties(data_folder=DATA_FOLDER)
    CATEGORY_PARENT_MAP = load_category_tree(data_folder=DATA_FOLDER)
    
    print("Calculating popular items for cold-start users...")
    POPULAR_ITEMS = get_popular_items(train_set, k=10)
    
    print("--- Artifacts loaded successfully. Ready to serve recommendations. ---")

def get_recommendations(visitor_id_str):
    """
    The main prediction function for the Gradio interface.
    
    Takes a visitor ID string, gets recommendations, and formats them for display.
    """
    try:
        visitor_id = int(visitor_id_str)
    except (ValueError, TypeError):
        return pd.DataFrame(), pd.DataFrame(), "Please enter a valid numerical Visitor ID."

    user_history_ids = DATAMODULE.user_history.get(visitor_id)

    def format_to_df(item_list):
        data = []
        for rank, item_id in enumerate(item_list, 1):
            category_id = ITEM_CATEGORY_MAP.get(item_id, 'N/A')
            parent_id = CATEGORY_PARENT_MAP.get(category_id, 'N/A') if pd.notna(category_id) else 'N/A'
            data.append([rank, item_id, category_id, parent_id])
        return pd.DataFrame(data, columns=['Rank', 'Item ID', 'Category ID', 'Parent ID'])

    # --- Cold-Start User (Fallback to Popularity) ---
    if user_history_ids is None:
        history_df = pd.DataFrame(columns=['Rank', 'Item ID', 'Category ID', 'Parent ID'])
        recs_df = format_to_df(POPULAR_ITEMS)
        message = f"User {visitor_id} is new. Showing Top 10 popular items as a fallback."
        return history_df, recs_df, message

    # --- Existing User (Use SASRec Model) ---
    history_df = format_to_df(user_history_ids)
    
    history_indices = [DATAMODULE.item_map[i] for i in user_history_ids if i in DATAMODULE.item_map]
    
    if not history_indices:
        message = "None of this user's historical items are in the model's vocabulary."
        return history_df, pd.DataFrame(), message

    max_len = DATAMODULE.max_len
    input_seq = history_indices[-max_len:]
    padded_input = np.zeros(max_len, dtype=np.int64)
    padded_input[-len(input_seq):] = input_seq
    
    input_tensor = torch.LongTensor(np.array([padded_input]))
    input_tensor = input_tensor.to(MODEL.device)

    with torch.no_grad():
        logits = MODEL(input_tensor)
        last_item_logits = logits[0, -1, :]
        top_indices = torch.topk(last_item_logits, 10).indices.tolist()

    recommended_item_ids = [DATAMODULE.inverse_item_map[idx] for idx in top_indices if idx in DATAMODULE.inverse_item_map]
    recs_df = format_to_df(recommended_item_ids)
    message = f"Showing personalized SASRec recommendations for user {visitor_id}."
    
    return history_df, recs_df, message

# --- Main Execution Block ---
if __name__ == "__main__":
    # Load all artifacts once at startup
    load_artifacts()

    # Find some valid example users to show in the UI
    users_in_train_history = set(DATAMODULE.user_history.keys())
    users_in_test_set = set(DATAMODULE.test_df['visitorid'].unique())
    valid_example_users = list(users_in_train_history.intersection(users_in_test_set))
    
    # Convert numpy types to standard Python int for Gradio compatibility
    example_list = [int(u) for u in valid_example_users[:4]] + [-999]

    # Create and launch the Gradio interface
    with gr.Blocks(theme=gr.themes.Soft(), title="SASRec Recommender") as iface:
        gr.Markdown(
            """
            # SASRec Sequential Recommender System
            An interactive demo of a state-of-the-art recommender system trained on the RetailRocket dataset.
            """
        )
        with gr.Row():
            with gr.Column(scale=1):
                visitor_id_input = gr.Number(
                    label="Enter Visitor ID", 
                    info="Enter a user's numerical ID to get recommendations."
                )
                submit_button = gr.Button("Get Recommendations", variant="primary")
                gr.Examples(
                    examples=example_list,
                    inputs=visitor_id_input,
                    label="Example User IDs (Click to try)"
                )
            
            with gr.Column(scale=3):
                status_message = gr.Textbox(label="Status", interactive=False)
                with gr.Tabs():
                    with gr.TabItem("Top 10 Recommendations"):
                        recs_output = gr.DataFrame(label="Recommended Items")
                    with gr.TabItem("User's Recent History"):
                        history_output = gr.DataFrame(label="Interaction History")
            
        submit_button.click(
            fn=get_recommendations,
            inputs=visitor_id_input,
            outputs=[history_output, recs_output, status_message]
        )

    # For local testing, this creates a shareable link.
    # On Hugging Face Spaces, this is not strictly necessary but doesn't hurt.
    iface.launch(share=True)