Update app.py

app.py

@@ -1,101 +1,123 @@
 import gradio as gr
-import torch
-from transformers import AutoTokenizer, AutoModelForSequenceClassification
+from transformers import pipeline
+from huggingface_hub import login
 import os
 
-# Load model and tokenizer
-MODEL_NAME = "mental/mental-longformer-depression"
-tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-model.to(device)
-model.eval()
-
-def predict_depression(text):
-    """Predict depression risk from user text"""
+# Authenticate with HuggingFace for private model access
+token = os.getenv("HF_TOKEN")
+if token:
+    login(token=token)
+
+# Load your private model (CORRECT MODEL NAME!)
+model_name = "avtak/erisk-longformer-depression-v1"
+classifier = pipeline("text-classification", model=model_name, truncation=True, max_length=4096, top_k=None)
+
+def get_risk_interpretation(depression_score):
+    """
+    Interprets depression risk score using research-based thresholds.
+    Optimal threshold from training: 0.6
+    """
+    if depression_score < 0.30:
+        return {
+            "level": "Low Risk",
+            "color": "🟢",
+            "interpretation": "Linguistic patterns consistent with non-depressed users. No significant markers of depression detected.",
+            "recommendation": "Continue monitoring mental wellness. Maintain healthy habits."
+        }
+    elif 0.30 <= depression_score < 0.60:
+        return {
+            "level": "Moderate Risk",
+            "color": "🟡",
+            "interpretation": "Some linguistic markers associated with depression detected. May indicate early signs or temporary mood changes.",
+            "recommendation": "Consider monitoring over time. Seek support if feelings persist. Talk to someone you trust."
+        }
+    else:  # >= 0.60 (optimal threshold from training)
+        return {
+            "level": "High Risk",
+            "color": "🔴",
+            "interpretation": "Strong linguistic patterns associated with depression detected. Multiple markers of depressive language present.",
+            "recommendation": "Strongly recommend speaking with a mental health professional. Immediate support resources provided below."
+        }
+
+def analyze_text(user_text):
+    """
+    Analyzes text for linguistic markers of depression using Mental-Longformer.
     
-    # Check text length
-    word_count = len(text.split())
+    This MCP tool provides depression risk assessment based on long-form user histories.
+    Trained on eRisk datasets (2017-2022) with F1-score of 0.7668.
+    Optimal decision threshold: 0.6
+    """
+    if not user_text.strip():
+        return "⚠️ Please provide text to analyze."
     
-    if word_count < 88:
-        warning = f"⚠️ Warning: Text Too Short ({word_count} words)"
-        message = (
-            "This model requires extensive text (300+ words recommended) for accurate analysis. "
-            "Short inputs will produce unreliable results.\n\n"
-            "Please provide:\n"
-            "• Multiple social media posts (10-15 posts)\n"
-            "• Extended writing samples\n"
-            "• Long-form personal narratives\n\n"
-            "Use the examples below to see proper input format."
-        )
-        return warning, message, ""
+    # Check text length
+    word_count = len(user_text.split())
+    if word_count < 100:
+        return f"""⚠️ **Warning: Text Too Short ({word_count} words)**
+
+This model requires extensive text (300+ words recommended) for accurate analysis. 
+Short inputs will produce unreliable results.
+
+Please provide:
+- Multiple social media posts (10-15 posts)
+- Extended writing samples
+- Long-form personal narratives
+
+Use the examples below to see proper input format."""
     
-    # Tokenize and predict
-    inputs = tokenizer(
-        text,
-        return_tensors="pt",
-        truncation=True,
-        max_length=4096,
-        padding=True
-    ).to(device)
+    results = classifier(user_text)[0]
     
-    with torch.no_grad():
-        outputs = model(**inputs)
-        probs = torch.nn.functional.softmax(outputs.logits, dim=-1)
-        depression_prob = probs[0][1].item() * 100
+    # Get depression probability (LABEL_1)
+    depression_prob = next((r['score'] for r in results if r['label'] == 'LABEL_1'), 0)
+    risk_info = get_risk_interpretation(depression_prob)
     
-    # Determine risk level
-    if depression_prob >= 60:
-        risk_level = "🔴 Depression Risk Assessment: High Risk"
-        risk_color = "#ff4444"
-        interpretation = (
-            "Strong linguistic patterns associated with depression detected. "
-            "Multiple markers of depressive language present."
-        )
-        recommendation = (
-            "Strongly recommend speaking with a mental health professional. "
-            "Immediate support resources provided below."
-        )
-    elif depression_prob >= 30:
-        risk_level = "🟡 Depression Risk Assessment: Moderate Risk"
-        risk_color = "#ffaa00"
-        interpretation = (
-            "Some concerning patterns detected. Monitoring recommended."
-        )
-        recommendation = (
-            "Consider reaching out to a mental health professional for guidance. "
-            "Early intervention can be helpful."
-        )
-    else:
-        risk_level = "🟢 Depression Risk Assessment: Low Risk"
-        risk_color = "#44ff44"
-        interpretation = (
-            "Typical language patterns, no significant depression markers detected."
-        )
-        recommendation = (
-            "Continue maintaining mental wellness. If concerns arise, "
-            "don't hesitate to seek professional support."
-        )
-    
-    # Format output
-    output = f"""
-{risk_level}
-Depression Probability: {depression_prob:.1f}%
-Text Length: {word_count} words
+    output = f"""# {risk_info['color']} Depression Risk Assessment: {risk_info['level']}
+
+**Depression Probability:** {depression_prob*100:.1f}%  
+**Text Length:** {word_count} words
+
+---
 
 ## Risk Interpretation
 
-{interpretation}
+**{risk_info['interpretation']}**
+
+### Recommended Action
+{risk_info['recommendation']}
+
+---
+
+## Model Technical Details
+- **Score Breakdown:** Depression: {depression_prob*100:.1f}% | Non-Depression: {(1-depression_prob)*100:.1f}%
+- **Decision Threshold:** 60% (optimal from training)
+- **Model Context:** 4,096 tokens
+- **Validation F1-Score:** 0.7668
+
+---
 
-## Recommended Action
+## ⚠️ Important Disclaimers
+
+**This is NOT a diagnostic tool.** This model:
+
+- Detects statistical patterns in language, not clinical depression
+- Requires professional interpretation
+- Cannot replace mental health assessment by qualified professionals
+- May produce false positives/negatives
 
-{recommendation}
+**If you're in crisis, help is available NOW:**
+
+- 🆘 **Crisis Text Line:** Text HOME to 741741 (US)
+- 📞 **National Suicide Prevention Lifeline:** 988 (US)
+- 🌍 **International:** [befrienders.org](https://befrienders.org)
+
+---
+
+*Analysis based on linguistic patterns from eRisk datasets. Consult healthcare professionals for mental health concerns.*
 """
-    
-    return risk_level, output, f"Depression Probability: {depression_prob:.1f}%"
+    return output
 
-# Example texts (no post numbers or separators)
-LOW_RISK_EXAMPLE = """I'm thinking of expanding my garden to include some tomatoes next. I've been watching a bunch of videos on how to build a small planter box. Feeling excited about the new project.
+# Clean examples WITHOUT post numbers or separators
+example_low_risk = """I'm thinking of expanding my garden to include some tomatoes next. I've been watching a bunch of videos on how to build a small planter box. Feeling excited about the new project.
 
 It's been a good week. Had some small wins, spent time with people I care about, and got to enjoy my hobby. Feeling content and looking forward to what next week brings.
 
@@ -107,7 +129,7 @@ Finally tried that new recipe I've been meaning to make. It turned out better th
 
 Work has been busy but manageable. I completed a challenging project ahead of schedule and my manager gave me positive feedback. Feeling accomplished and motivated."""
 
-MODERATE_RISK_EXAMPLE = """Been feeling a bit off lately. Not sure what it is, but I just don't have the same energy I used to have. Maybe I'm just tired from work.
+example_moderate_risk = """Been feeling a bit off lately. Not sure what it is, but I just don't have the same energy I used to have. Maybe I'm just tired from work.
 
 My sleep schedule has been all over the place. Some nights I can't fall asleep, other nights I sleep too much. Either way, I wake up feeling drained.
 
@@ -119,7 +141,7 @@ I've been spending a lot of time scrolling through my phone. I know it's not pro
 
 Haven't been eating as healthy as I should. Skipping meals or just grabbing whatever is convenient. Cooking feels like too much effort right now."""
 
-HIGH_RISK_EXAMPLE = """I woke up feeling exhausted even after sleeping for 10 hours. The thought of getting out of bed feels overwhelming. I don't even know why I bother anymore.
+example_high_risk = """I woke up feeling exhausted even after sleeping for 10 hours. The thought of getting out of bed feels overwhelming. I don't even know why I bother anymore.
 
 My apartment is a mess. I know I should clean it but I just can't find the energy or motivation. I look around and it just makes me feel worse, like the mess on the outside matches the mess on the inside.
 
@@ -133,7 +155,7 @@ I feel like I'm drowning in quicksand. The more I struggle, the faster I sink. W
 
 I just want it to stop. The noise in my head. The constant feeling of dread. I'm so tired of fighting. I just want to feel okay again, but I've forgotten what that even feels like."""
 
-# Build Gradio interface
+# Build Gradio interface with examples at TOP
 with gr.Blocks(theme=gr.themes.Soft(), title="Early Depression Detection") as demo:
     
     # Header
@@ -155,42 +177,42 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Early Depression Detection") as de
     )
     
     with gr.Row():
-        gr.Markdown("[📄 Model Card](https://huggingface.co/mental/mental-longformer-depression)")
-        gr.Markdown("[👤 Author LinkedIn](https://www.linkedin.com/in/yourprofile)")
+        gr.Markdown("[📄 Model Card](https://huggingface.co/avtak/erisk-longformer-depression-v1)")
+        gr.Markdown("[👤 Author LinkedIn](https://www.linkedin.com/in/hassanzh/)")
     
-    # Examples Section at Top
+    # Examples Section at Top (BEFORE input textbox)
     gr.Markdown("## 📋 Example Inputs")
     gr.Markdown(
         "Click an example below to see how the model works. "
         "Each category shows different risk levels based on linguistic patterns."
     )
     
-    with gr.Row():
-        user_input = gr.Textbox(
-            label="📝 Enter User Text History (300+ words recommended)",
-            placeholder="Paste multiple social media posts, journal entries, or extended writing samples here...",
-            lines=15,
-            max_lines=20
-        )
+    # Create the input textbox first (but it will be referenced by examples)
+    user_input = gr.Textbox(
+        label="📝 Enter User Text History (300+ words recommended)",
+        placeholder="Paste multiple social media posts, journal entries, or extended writing samples here...",
+        lines=15,
+        max_lines=20
+    )
     
-    # Three categories of examples
+    # Three categories of examples ABOVE the textbox visually
     with gr.Accordion("🟢 Low Risk Examples", open=False):
         gr.Examples(
-            examples=[[LOW_RISK_EXAMPLE]],
+            examples=[[example_low_risk]],
             inputs=user_input,
             label="Positive, engaged content"
         )
     
     with gr.Accordion("🟡 Moderate Risk Examples", open=False):
         gr.Examples(
-            examples=[[MODERATE_RISK_EXAMPLE]],
+            examples=[[example_moderate_risk]],
             inputs=user_input,
             label="Some concerning patterns"
         )
     
     with gr.Accordion("🔴 High Risk Examples", open=True):
         gr.Examples(
-            examples=[[HIGH_RISK_EXAMPLE]],
+            examples=[[example_high_risk]],
             inputs=user_input,
             label="Strong depression markers"
         )
@@ -201,20 +223,10 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Early Depression Detection") as de
         submit_btn = gr.Button("Submit", variant="primary", size="lg")
     
     # Output Section
-    risk_title = gr.Markdown("## Assessment Result")
-    output_text = gr.Markdown()
+    output_text = gr.Markdown(label="📊 Risk Assessment Results")
     
     # Model Technical Details
-    with gr.Accordion("📊 Model Technical Details", open=False):
-        gr.Markdown("""
-- **Score Breakdown:** Depression: 89.8% | Non-Depression: 10.2%
-- **Decision Threshold:** 60% (optimal from training)
-- **Model Context:** 4,096 tokens
-- **Validation F1-Score:** 0.7668
-        """)
-    
-    # Risk Scale Info
-    with gr.Accordion("ℹ️ About the Risk Scale", open=False):
+    with gr.Accordion("📊 About the Risk Scale", open=False):
         gr.Markdown("""
 This model uses research-validated thresholds:
 
@@ -234,20 +246,11 @@ This tool is for **research and screening purposes only**. It should never be us
 - To make clinical decisions without human oversight
 - For surveillance or discrimination
 - Without informed consent
-
-**May produce false positives/negatives**
         """)
     
     # Important Disclaimers
-    gr.Markdown("## ⚠️ Important Disclaimers")
+    gr.Markdown("## ⚠️ Crisis Resources")
     gr.Markdown("""
-**This is NOT a diagnostic tool.** This model:
-
-- Detects statistical patterns in language, not clinical depression
-- Requires professional interpretation
-- Cannot replace mental health assessment by qualified professionals
-- May produce false positives/negatives
-
 **If you're in crisis, help is available NOW:**
 
 - 🆘 **Crisis Text Line:** Text HOME to 741741 (US)
@@ -255,27 +258,19 @@ This tool is for **research and screening purposes only**. It should never be us
 - 🌍 **International:** [befrienders.org](https://www.befrienders.org)
     """)
     
-    gr.Markdown(
-        "*Analysis based on linguistic patterns from eRisk datasets. "
-        "Consult healthcare professionals for mental health concerns.*"
-    )
-    
-    with gr.Row():
-        gr.Button("🔗 Share via Link", link="https://huggingface.co/spaces/avtak/depression-detection-mcp-agent")
-    
     # Event handlers
     submit_btn.click(
-        fn=predict_depression,
+        fn=analyze_text,
         inputs=user_input,
-        outputs=[risk_title, output_text, gr.Textbox(visible=False)]
+        outputs=output_text
     )
     
     clear_btn.click(
-        fn=lambda: ("", "## Assessment Result", ""),
+        fn=lambda: "",
         inputs=None,
-        outputs=[user_input, risk_title, output_text]
+        outputs=user_input
     )
 
-# Launch
+# 🔥 Enable MCP Server
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(mcp_server=True)