Spaces:
Sleeping
Sleeping
Update model.py
Browse files
model.py
CHANGED
@@ -6,7 +6,7 @@ import gradio as gr
|
|
6 |
from scipy.sparse import csr_matrix
|
7 |
|
8 |
class MatrixFactorization:
|
9 |
-
def __init__(self, n_factors=50):
|
10 |
self.n_factors = n_factors
|
11 |
self.model = TruncatedSVD(n_components=n_factors, random_state=42)
|
12 |
self.user_title_matrix = None
|
@@ -18,16 +18,12 @@ class MatrixFactorization:
|
|
18 |
print("Training model...")
|
19 |
start_time = time.time()
|
20 |
|
21 |
-
# Get top songs
|
22 |
-
top_songs =
|
23 |
-
.sum()
|
24 |
-
.sort_values(ascending=False)
|
25 |
-
.head(10000)
|
26 |
-
.index)
|
27 |
-
|
28 |
df_filtered = df[df['title'].isin(top_songs)]
|
29 |
-
print(
|
30 |
|
|
|
31 |
pivot = pd.pivot_table(
|
32 |
df_filtered,
|
33 |
values='play_count',
|
@@ -36,8 +32,11 @@ class MatrixFactorization:
|
|
36 |
fill_value=0
|
37 |
)
|
38 |
self.column_names = pivot.columns
|
|
|
|
|
39 |
self.user_title_matrix = csr_matrix(pivot.values)
|
40 |
|
|
|
41 |
self.titles_df = df_filtered.groupby('title').agg({
|
42 |
'artist_name': 'first',
|
43 |
'year': 'first',
|
@@ -49,13 +48,45 @@ class MatrixFactorization:
|
|
49 |
self.user_vectors = self.model.fit_transform(self.user_title_matrix)
|
50 |
self.item_vectors = self.model.components_
|
51 |
|
52 |
-
#
|
53 |
-
self._cached_choices = self.
|
54 |
|
55 |
print(f"Training completed in {time.time() - start_time:.2f} seconds")
|
56 |
|
57 |
-
def
|
58 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
for title, row in self.titles_df.iterrows():
|
60 |
display_text = f"{title} β’ by {row['artist_name']}"
|
61 |
extra_info = []
|
@@ -65,11 +96,8 @@ class MatrixFactorization:
|
|
65 |
extra_info.append(str(int(row['year'])))
|
66 |
if extra_info:
|
67 |
display_text += f" [{', '.join(extra_info)}]"
|
68 |
-
|
69 |
-
return sorted(
|
70 |
-
|
71 |
-
def create_title_choices(self):
|
72 |
-
return self._cached_choices if self._cached_choices else self._generate_choices()
|
73 |
|
74 |
def create_gradio_interface(mf_model):
|
75 |
try:
|
@@ -77,7 +105,7 @@ def create_gradio_interface(mf_model):
|
|
77 |
gr.Markdown("""# π΅ Music Recommendation System πΆ
|
78 |
|
79 |
### Instructions:
|
80 |
-
1. β³
|
81 |
2. π Search by title, artist, album, or year
|
82 |
3. π§ Select up to 5 songs
|
83 |
4. π Click for recommendations
|
@@ -86,7 +114,7 @@ def create_gradio_interface(mf_model):
|
|
86 |
|
87 |
with gr.Row():
|
88 |
input_songs = gr.Dropdown(
|
89 |
-
choices=mf_model.
|
90 |
label="Search and select songs (up to 5)",
|
91 |
info="Format: Title β’ by Artist [Album, Year]",
|
92 |
multiselect=True,
|
@@ -102,7 +130,7 @@ def create_gradio_interface(mf_model):
|
|
102 |
)
|
103 |
|
104 |
recommend_btn.click(
|
105 |
-
fn=mf_model.
|
106 |
inputs=input_songs,
|
107 |
outputs=output_table
|
108 |
)
|
|
|
6 |
from scipy.sparse import csr_matrix
|
7 |
|
8 |
class MatrixFactorization:
|
9 |
+
def __init__(self, n_factors=50):
|
10 |
self.n_factors = n_factors
|
11 |
self.model = TruncatedSVD(n_components=n_factors, random_state=42)
|
12 |
self.user_title_matrix = None
|
|
|
18 |
print("Training model...")
|
19 |
start_time = time.time()
|
20 |
|
21 |
+
# Get top 10000 songs
|
22 |
+
top_songs = df.groupby('title')['play_count'].sum().nlargest(10000).index
|
|
|
|
|
|
|
|
|
|
|
23 |
df_filtered = df[df['title'].isin(top_songs)]
|
24 |
+
print("Filtered to 10000 most played songs")
|
25 |
|
26 |
+
# Create pivot table
|
27 |
pivot = pd.pivot_table(
|
28 |
df_filtered,
|
29 |
values='play_count',
|
|
|
32 |
fill_value=0
|
33 |
)
|
34 |
self.column_names = pivot.columns
|
35 |
+
|
36 |
+
# Convert to sparse matrix
|
37 |
self.user_title_matrix = csr_matrix(pivot.values)
|
38 |
|
39 |
+
# Create titles dataframe
|
40 |
self.titles_df = df_filtered.groupby('title').agg({
|
41 |
'artist_name': 'first',
|
42 |
'year': 'first',
|
|
|
48 |
self.user_vectors = self.model.fit_transform(self.user_title_matrix)
|
49 |
self.item_vectors = self.model.components_
|
50 |
|
51 |
+
# Cache choices
|
52 |
+
self._cached_choices = self.create_title_choices()
|
53 |
|
54 |
print(f"Training completed in {time.time() - start_time:.2f} seconds")
|
55 |
|
56 |
+
def get_recommendations(self, selected_titles):
|
57 |
+
if not selected_titles:
|
58 |
+
return []
|
59 |
+
|
60 |
+
try:
|
61 |
+
actual_titles = [title.split(" β’ by ")[0] for title in selected_titles]
|
62 |
+
title_to_idx = {title: idx for idx, title in enumerate(self.column_names)}
|
63 |
+
selected_indices = [title_to_idx[title] for title in actual_titles]
|
64 |
+
|
65 |
+
user_vector = np.mean([self.item_vectors[:, idx] for idx in selected_indices], axis=0)
|
66 |
+
scores = np.dot(user_vector, self.item_vectors)
|
67 |
+
|
68 |
+
title_scores = [(title, score) for title, score in zip(self.column_names, scores)
|
69 |
+
if title not in actual_titles]
|
70 |
+
recommendations = sorted(title_scores, key=lambda x: x[1], reverse=True)[:5]
|
71 |
+
|
72 |
+
results = []
|
73 |
+
for title, score in recommendations:
|
74 |
+
row = self.titles_df.loc[title]
|
75 |
+
confidence = 30 + (score * 70)
|
76 |
+
results.append([
|
77 |
+
title,
|
78 |
+
row['artist_name'],
|
79 |
+
int(row['year']) if pd.notna(row['year']) else None,
|
80 |
+
f"{min(max(confidence, 30), 100):.2f}%"
|
81 |
+
])
|
82 |
+
return results
|
83 |
+
|
84 |
+
except Exception as e:
|
85 |
+
print(f"Error in recommendations: {str(e)}")
|
86 |
+
return []
|
87 |
+
|
88 |
+
def create_title_choices(self):
|
89 |
+
title_choices = []
|
90 |
for title, row in self.titles_df.iterrows():
|
91 |
display_text = f"{title} β’ by {row['artist_name']}"
|
92 |
extra_info = []
|
|
|
96 |
extra_info.append(str(int(row['year'])))
|
97 |
if extra_info:
|
98 |
display_text += f" [{', '.join(extra_info)}]"
|
99 |
+
title_choices.append(display_text)
|
100 |
+
return sorted(title_choices)
|
|
|
|
|
|
|
101 |
|
102 |
def create_gradio_interface(mf_model):
|
103 |
try:
|
|
|
105 |
gr.Markdown("""# π΅ Music Recommendation System πΆ
|
106 |
|
107 |
### Instructions:
|
108 |
+
1. β³ Model loads top 10000 songs (~1 min)
|
109 |
2. π Search by title, artist, album, or year
|
110 |
3. π§ Select up to 5 songs
|
111 |
4. π Click for recommendations
|
|
|
114 |
|
115 |
with gr.Row():
|
116 |
input_songs = gr.Dropdown(
|
117 |
+
choices=mf_model._cached_choices,
|
118 |
label="Search and select songs (up to 5)",
|
119 |
info="Format: Title β’ by Artist [Album, Year]",
|
120 |
multiselect=True,
|
|
|
130 |
)
|
131 |
|
132 |
recommend_btn.click(
|
133 |
+
fn=mf_model.get_recommendations,
|
134 |
inputs=input_songs,
|
135 |
outputs=output_table
|
136 |
)
|