updated app.py

app.py

@@ -10,6 +10,7 @@ import gradio as gr
 from transformers import AutoTokenizer, AutoModel
 from sklearn.ensemble import RandomForestClassifier
 from sklearn.preprocessing import StandardScaler
+from sklearn.metrics import accuracy_score, f1_score
 
 # =========================
 # 1. DOWNLOAD DATASET
@@ -30,7 +31,7 @@ if not os.path.exists(extract_path):
         zip_ref.extractall(extract_path)
 
 # =========================
-# 2. LOAD BERT
+# 2. LOAD BERT MODEL
 # =========================
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
@@ -39,7 +40,7 @@ bert = AutoModel.from_pretrained("bert-base-multilingual-cased").to(device)
 bert.eval()
 
 # =========================
-# FEATURE FUNCTIONS
+# 3. FEATURE FUNCTIONS
 # =========================
 def get_text_embedding(text):
     inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=128).to(device)
@@ -68,19 +69,18 @@ def get_label(folder):
     return 1 if "_P" in folder else 0
 
 # =========================
-# 3. TRAIN MODEL (LIGHT)
+# 4. TRAIN MODEL (LIGHT)
 # =========================
 print("Training model...")
 
 X, y = [], []
-folders = os.listdir(extract_path)[:20]  # 🔥 only 20 samples
+
+folders = [f for f in os.listdir(extract_path) if os.path.isdir(os.path.join(extract_path, f))]
+folders = folders[:20]  # 🔥 small subset for speed
 
 for folder in folders:
     path = os.path.join(extract_path, folder)
 
-    if not os.path.isdir(path):
-        continue
-
     text = load_text(path)
     text_feat = get_text_embedding(text)
     audio_feat = get_audio_features(path)
@@ -96,44 +96,56 @@ y = np.array(y)
 scaler = StandardScaler()
 X = scaler.fit_transform(X)
 
-model = RandomForestClassifier(n_estimators=50)
+model = RandomForestClassifier(n_estimators=50, random_state=42)
 model.fit(X, y)
 
 print("Model ready!")
 
 # =========================
-# 4. PREDICTION FUNCTION
+# 5. RUN PREDICTIONS
 # =========================
-def predict(text, audio):
+def run_on_dataset():
 
-    text_feat = get_text_embedding(text)
+    results = []
+    preds = []
+    labels = []
+
+    for i, folder in enumerate(folders):
+        path = os.path.join(extract_path, folder)
+
+        text = load_text(path)
+        text_feat = get_text_embedding(text)
+        audio_feat = get_audio_features(path)
+
+        x = np.concatenate([text_feat, audio_feat])
+        x = scaler.transform([x])
+
+        pred = model.predict(x)[0]
+        label = get_label(folder)
+
+        preds.append(pred)
+        labels.append(label)
 
-    if audio:
-        y_audio, sr = librosa.load(audio, sr=16000)
-        mfcc = librosa.feature.mfcc(y=y_audio, sr=sr, n_mfcc=20)
-        audio_feat = np.mean(mfcc.T, axis=0)
-    else:
-        audio_feat = np.zeros(20)
+        results.append(f"{folder} → {'Depressed' if pred==1 else 'Control'}")
 
-    x = np.concatenate([text_feat, audio_feat])
-    x = scaler.transform([x])
+    acc = accuracy_score(labels, preds)
+    f1 = f1_score(labels, preds)
 
-    pred = model.predict(x)[0]
+    output = "\n".join(results)
+    output += f"\n\nAccuracy: {acc:.3f}"
+    output += f"\nF1 Score: {f1:.3f}"
 
-    return "Depression Detected ⚠️" if pred == 1 else "No Depression ✅"
+    return output
 
 # =========================
-# 5. UI
+# 6. GRADIO UI
 # =========================
 app = gr.Interface(
-    fn=predict,
-    inputs=[
-        gr.Textbox(label="Enter Text"),
-        gr.Audio(type="filepath", label="Upload Audio (optional)")
-    ],
+    fn=run_on_dataset,
+    inputs=[],
     outputs="text",
     title="Multimodal Depression Detection",
-    description="Training + inference demo using text and audio"
+    description="Automatically trains on dataset and evaluates predictions (text + audio)"
 )
 
 app.launch()