EnYa32 commited on
Commit
c09cd2f
·
verified ·
1 Parent(s): b79b18c

Update src/streamlit_app.py

Browse files
Files changed (1) hide show
  1. src/streamlit_app.py +115 -95
src/streamlit_app.py CHANGED
@@ -8,18 +8,16 @@ from pathlib import Path
8
  # Page config
9
  # -------------------------
10
  st.set_page_config(
11
- page_title='Clustering Predictor (KMeans / GMM)',
12
  page_icon='🧩',
13
  layout='centered'
14
  )
15
 
16
- st.title('🧩 Clustering Predictor (KMeans / GMM)')
17
- st.write(
18
- 'Upload a CSV file and get cluster predictions using your saved preprocessing (Scaler + PCA) and model.'
19
- )
20
 
21
  # -------------------------
22
- # Paths (HF friendly)
23
  # -------------------------
24
  BASE_DIR = Path(__file__).resolve().parent
25
 
@@ -35,15 +33,11 @@ GMM_PATH = BASE_DIR / 'gmm_model.pkl'
35
  # -------------------------
36
  @st.cache_resource
37
  def load_assets():
38
- missing = []
39
- for p in [FEATURES_PATH, SCALER_PATH, PCA_PATH]:
40
- if not p.exists():
41
- missing.append(p.name)
42
-
43
  if missing:
44
  raise FileNotFoundError(
45
- f'Missing required files in repo root: {missing}. '
46
- 'Please upload them next to app.py.'
47
  )
48
 
49
  feature_names = joblib.load(FEATURES_PATH)
@@ -51,14 +45,14 @@ def load_assets():
51
  pca = joblib.load(PCA_PATH)
52
 
53
  models = {}
54
- if KMEANS_PATH.exists():
55
- models['KMeans (k=9)'] = joblib.load(KMEANS_PATH)
56
  if GMM_PATH.exists():
57
- models['GMM (k=9)'] = joblib.load(GMM_PATH)
 
 
58
 
59
  if not models:
60
  raise FileNotFoundError(
61
- "No model files found. Upload 'kmeans_model_k9.pkl' and/or 'gmm_model_k9.pkl' next to app.py."
62
  )
63
 
64
  return feature_names, scaler, pca, models
@@ -70,92 +64,118 @@ except Exception as e:
70
  st.stop()
71
 
72
  # -------------------------
73
- # Model selector
74
  # -------------------------
75
- model_name = st.selectbox('Select model', list(models.keys()))
76
  model = models[model_name]
77
 
78
- st.caption('Expected input columns:')
79
- with st.expander('Show feature columns'):
80
- st.write(feature_names)
81
-
82
- # -------------------------
83
- # Upload CSV
84
- # -------------------------
85
- uploaded = st.file_uploader('Upload CSV', type=['csv'])
86
-
87
- def preprocess_df(df_in: pd.DataFrame) -> tuple[pd.DataFrame, pd.Series | None]:
88
- # Accept either 'id' or 'Id' columns (optional)
89
- id_col = None
90
- if 'id' in df_in.columns:
91
- id_col = 'id'
92
- elif 'Id' in df_in.columns:
93
- id_col = 'Id'
94
-
95
- ids = df_in[id_col].copy() if id_col else None
96
-
97
- X = df_in.drop(columns=[id_col], errors='ignore').copy()
98
-
99
- # Validate columns
100
- missing_cols = [c for c in feature_names if c not in X.columns]
101
- extra_cols = [c for c in X.columns if c not in feature_names]
102
-
103
- if missing_cols:
104
- raise ValueError(f'Missing required columns: {missing_cols}')
105
-
106
- # Keep only expected columns + correct order
107
- X = X[feature_names]
108
 
109
- # Convert to numeric if possible (safety)
110
- for c in X.columns:
111
- X[c] = pd.to_numeric(X[c], errors='coerce')
112
 
113
- if X.isna().any().any():
114
- # You can choose a strategy; here we fail fast so user fixes the input.
115
- bad_cols = X.columns[X.isna().any()].tolist()
116
- raise ValueError(
117
- f'Found NaNs after converting to numeric. Check these columns: {bad_cols}. '
118
- 'Make sure your CSV has valid numeric values.'
119
- )
120
-
121
- return X, ids
122
 
123
- def predict_clusters(X: pd.DataFrame) -> np.ndarray:
124
- X_scaled = scaler.transform(X)
125
  X_pca = pca.transform(X_scaled)
126
-
127
- # KMeans and GMM both have predict()
128
  preds = model.predict(X_pca)
129
  return preds
130
 
131
- if uploaded is not None:
132
- try:
133
- df_up = pd.read_csv(uploaded)
134
- X_up, ids = preprocess_df(df_up)
135
- preds = predict_clusters(X_up)
136
-
137
- out = pd.DataFrame({'Predicted': preds})
138
- if ids is not None:
139
- out.insert(0, 'Id', ids)
140
-
141
- st.success('✅ Predictions created successfully.')
142
- st.dataframe(out.head(30), use_container_width=True)
143
-
144
- st.download_button(
145
- 'Download predictions as CSV',
146
- data=out.to_csv(index=False).encode('utf-8'),
147
- file_name='predictions.csv',
148
- mime='text/csv'
149
- )
150
-
151
- # Quick info
152
- st.subheader('Cluster distribution')
153
- dist = pd.Series(preds).value_counts().sort_index()
154
- st.write(dist)
155
-
156
- except Exception as e:
157
- st.error(str(e))
158
- st.stop()
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
159
 
160
- else:
161
- st.info('Upload a CSV file to generate cluster predictions.')
 
 
 
 
8
  # Page config
9
  # -------------------------
10
  st.set_page_config(
11
+ page_title='Clustering Predictor (GMM / KMeans)',
12
  page_icon='🧩',
13
  layout='centered'
14
  )
15
 
16
+ st.title('🧩 Clustering Predictor')
17
+ st.write('Predict cluster labels using saved preprocessing (Scaler + PCA) and a clustering model.')
 
 
18
 
19
  # -------------------------
20
+ # Paths
21
  # -------------------------
22
  BASE_DIR = Path(__file__).resolve().parent
23
 
 
33
  # -------------------------
34
  @st.cache_resource
35
  def load_assets():
36
+ required = [FEATURES_PATH, SCALER_PATH, PCA_PATH]
37
+ missing = [p.name for p in required if not p.exists()]
 
 
 
38
  if missing:
39
  raise FileNotFoundError(
40
+ f'Missing required files in repo root: {missing}. Put them next to app.py.'
 
41
  )
42
 
43
  feature_names = joblib.load(FEATURES_PATH)
 
45
  pca = joblib.load(PCA_PATH)
46
 
47
  models = {}
 
 
48
  if GMM_PATH.exists():
49
+ models['GMM (best)'] = joblib.load(GMM_PATH)
50
+ if KMEANS_PATH.exists():
51
+ models['KMeans'] = joblib.load(KMEANS_PATH)
52
 
53
  if not models:
54
  raise FileNotFoundError(
55
+ "No model found. Upload 'gmm_model_k9.pkl' (and optionally 'kmeans_model_k9.pkl') next to app.py."
56
  )
57
 
58
  return feature_names, scaler, pca, models
 
64
  st.stop()
65
 
66
  # -------------------------
67
+ # Model selection
68
  # -------------------------
69
+ model_name = st.selectbox('Select model', list(models.keys()), index=0)
70
  model = models[model_name]
71
 
72
+ def predict_from_features(df_features: pd.DataFrame) -> np.ndarray:
73
+ # Ensure correct order
74
+ df_features = df_features[feature_names].copy()
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
75
 
76
+ # Convert to numeric
77
+ for c in df_features.columns:
78
+ df_features[c] = pd.to_numeric(df_features[c], errors='coerce')
79
 
80
+ if df_features.isna().any().any():
81
+ bad_cols = df_features.columns[df_features.isna().any()].tolist()
82
+ raise ValueError(f'NaN values found in columns: {bad_cols}. Please provide valid numeric values.')
 
 
 
 
 
 
83
 
84
+ X_scaled = scaler.transform(df_features)
 
85
  X_pca = pca.transform(X_scaled)
 
 
86
  preds = model.predict(X_pca)
87
  return preds
88
 
89
+ # -------------------------
90
+ # UI Tabs
91
+ # -------------------------
92
+ tab_csv, tab_single = st.tabs(['📁 CSV Upload', '🧮 Single Prediction'])
93
+
94
+ # =========================
95
+ # Tab 1: CSV Upload
96
+ # =========================
97
+ with tab_csv:
98
+ st.subheader('Upload a CSV and predict clusters')
99
+ uploaded = st.file_uploader('Upload CSV', type=['csv'])
100
+
101
+ if uploaded is not None:
102
+ try:
103
+ df_up = pd.read_csv(uploaded)
104
+
105
+ # id handling (optional)
106
+ id_col = None
107
+ if 'id' in df_up.columns:
108
+ id_col = 'id'
109
+ elif 'Id' in df_up.columns:
110
+ id_col = 'Id'
111
+
112
+ ids = df_up[id_col].copy() if id_col else None
113
+ X_up = df_up.drop(columns=[id_col], errors='ignore')
114
+
115
+ missing_cols = [c for c in feature_names if c not in X_up.columns]
116
+ if missing_cols:
117
+ st.error(f'Missing required columns: {missing_cols}')
118
+ st.stop()
119
+
120
+ X_up = X_up[feature_names]
121
+ preds = predict_from_features(X_up)
122
+
123
+ out = pd.DataFrame({'Predicted': preds})
124
+ if ids is not None:
125
+ out.insert(0, 'Id', ids)
126
+
127
+ st.success('✅ Predictions created successfully.')
128
+ st.dataframe(out.head(30), use_container_width=True)
129
+
130
+ st.download_button(
131
+ 'Download predictions as CSV',
132
+ data=out.to_csv(index=False).encode('utf-8'),
133
+ file_name='predictions.csv',
134
+ mime='text/csv'
135
+ )
136
+
137
+ st.subheader('Cluster distribution')
138
+ st.write(pd.Series(preds).value_counts().sort_index())
139
+
140
+ except Exception as e:
141
+ st.error(str(e))
142
+
143
+ else:
144
+ st.info('Upload a CSV file to generate cluster predictions.')
145
+
146
+ # =========================
147
+ # Tab 2: Single Prediction
148
+ # =========================
149
+ with tab_single:
150
+ st.subheader('Enter one row of features and predict its cluster')
151
+
152
+ # Optional: user-friendly note
153
+ st.caption('Tip: If you don’t know values, use a row from your dataset as an example.')
154
+
155
+ # Build form inputs dynamically
156
+ with st.form('single_pred_form'):
157
+ cols = st.columns(2)
158
+ values = {}
159
+
160
+ for i, feat in enumerate(feature_names):
161
+ # Default 0.0 is safe; you can also set dataset mean if you want
162
+ if i % 2 == 0:
163
+ values[feat] = cols[0].number_input(feat, value=0.0)
164
+ else:
165
+ values[feat] = cols[1].number_input(feat, value=0.0)
166
+
167
+ submitted = st.form_submit_button('Predict cluster')
168
+
169
+ if submitted:
170
+ try:
171
+ one_row = pd.DataFrame([values], columns=feature_names)
172
+ pred = predict_from_features(one_row)[0]
173
+ st.success(f'✅ Predicted cluster: **{int(pred)}**')
174
+ except Exception as e:
175
+ st.error(str(e))
176
 
177
+ # -------------------------
178
+ # Footer
179
+ # -------------------------
180
+ with st.expander('Show expected feature columns'):
181
+ st.write(feature_names)