Update app.py

app.py

@@ -23,23 +23,6 @@ import spaces
 import requests
 from PIL import Image
 from transformers import pipeline
-from pathlib import Path
-
-def data_badge(content):
-    """Render a compact informational badge for data/model disclosures."""
-    return f"""
-    <div style="
-        padding: 12px 16px;
-        margin: 8px 0 16px 0;
-        border-radius: 10px;
-        border: 1px solid rgba(120, 130, 150, 0.25);
-        background: rgba(120, 130, 150, 0.08);
-        font-size: 0.9em;
-        line-height: 1.5;
-    ">
-        {content}
-    </div>
-    """
 
 # ---------------------------------------------------------------------------
 # Config — override any of these via Space "Variables and secrets"
@@ -54,7 +37,9 @@ HF_TOKEN = os.environ.get("HF_TOKEN")  # add as a Space SECRET to enable the cha
 ROUTER_URL = "https://router.huggingface.co/v1/chat/completions"
 
 PAGE_DISCLAIMER = (
-    "This tool is a research/portfolio project, not a medical device."
+    "This tool is a research/portfolio project, not a medical device. It hasn't "
+    "been clinically validated. Use it to get informed and organize your thoughts "
+    "before a real appointment — not as a substitute for one."
 )
 
 # Optional second classifier specifically for pigmented-lesion cases (HAM10000-based,
@@ -191,7 +176,15 @@ def chat_completion(messages, max_tokens=500):
         json={"model": CHAT_MODEL_ID, "messages": messages, "max_tokens": max_tokens},
         timeout=60,
     )
-    resp.raise_for_status()
+    if not resp.ok:
+        # Surface the actual reason from the response body (e.g. "model not
+        # supported by any provider") instead of just the generic status line —
+        # that detail is what actually explains a 400/404, not "Bad Request".
+        try:
+            detail = resp.json()
+        except Exception:
+            detail = resp.text
+        raise RuntimeError(f"{resp.status_code} error from router for model '{CHAT_MODEL_ID}': {detail}")
     data = resp.json()
     return data["choices"][0]["message"]["content"]
 
@@ -380,35 +373,59 @@ def chat_respond(message, history):
 # ---------------------------------------------------------------------------
 # Model Performance tab data loading
 # ---------------------------------------------------------------------------
-BASE_DIR = Path(__file__).resolve().parent
-ASSETS_DIR = BASE_DIR / "assets"
-
-
-def load_performance_artifacts():
-    """Load model-evaluation images generated by the evaluation notebook."""
+ASSETS_DIR = "assets"
 
-    artifacts = {
-        "unified_confusion_matrix": None,
-        "melanoma_threshold_curve": None,
-        "cluster_confusion_matrix": None,
-    }
 
-    if not ASSETS_DIR.is_dir():
-        return artifacts
+def load_performance_reports():
+    """Scan assets/ for *_metrics.json files and pair each with its confusion-matrix
+    image if present. Returns a list of dicts; empty list if nothing's been uploaded yet."""
+    import json
 
-    files = {
-        "unified_confusion_matrix": "unified_confusion_matrix.png",
-        "melanoma_threshold_curve": "unified_melanoma_threshold_curve.png",
-        "cluster_confusion_matrix": "cluster_confusion_matrix.png",
-    }
-
-    for key, filename in files.items():
-        path = ASSETS_DIR / filename
+    reports = []
+    if not os.path.isdir(ASSETS_DIR):
+        return reports
+    for fname in sorted(os.listdir(ASSETS_DIR)):
+        if not fname.endswith("_metrics.json"):
+            continue
+        prefix = fname[: -len("_metrics.json")]
+        try:
+            with open(os.path.join(ASSETS_DIR, fname)) as f:
+                data = json.load(f)
+        except Exception:
+            continue
+        image_path = None
+        for ext in (".png", ".jpg", ".jpeg"):
+            candidate = os.path.join(ASSETS_DIR, f"{prefix}_confusion_matrix{ext}")
+            if os.path.isfile(candidate):
+                image_path = candidate
+                break
+        reports.append({"data": data, "image_path": image_path})
+    return reports
+
+
+def format_performance_markdown(report):
+    data = report["data"]
+    lines = [f"### {data.get('model_name', 'Model')}"]
+    acc = data.get("accuracy")
+    f1m = data.get("f1_macro")
+    if acc is not None:
+        lines.append(f"**Accuracy:** {acc*100:.1f}%  ·  **Macro F1:** {f1m*100:.1f}%" if f1m is not None else f"**Accuracy:** {acc*100:.1f}%")
+
+    per_class = data.get("per_class", {})
+    if per_class:
+        lines.append("\n| Class | Precision | Recall | F1 | Support |")
+        lines.append("|---|---|---|---|---|")
+        for cls, m in per_class.items():
+            lines.append(
+                f"| {cls} | {m.get('precision', 0)*100:.1f}% | {m.get('recall', 0)*100:.1f}% "
+                f"| {m.get('f1', 0)*100:.1f}% | {int(m.get('support', 0))} |"
+            )
 
-        if path.is_file():
-            artifacts[key] = str(path)
+    notes = data.get("notes")
+    if notes:
+        lines.append(f"\n_Note: {notes}_")
 
-    return artifacts
+    return "\n".join(lines)
 
 
 ABOUT_MARKDOWN = f"""
@@ -449,6 +466,18 @@ about what it isn't: a validated medical device.
 - 🟢 **Low** — typically manageable with general skin care. Doctor visit
   optional, especially if unsure or it's not improving.
 
+### Known limitations (read this before trusting the output)
+- **Not clinically validated.** Benchmark accuracy on a public dataset is
+  not the same as real-world diagnostic reliability.
+- **Dataset label limitations.** The DermNet taxonomy combines melanoma
+  with benign moles under one label — the app defaults that whole category
+  to High severity rather than pretending to distinguish them.
+- **Skin-tone representation.** Public dermatology datasets skew toward
+  lighter skin tones; performance on underrepresented skin tones is
+  unverified without a dedicated bias audit.
+- **No clinical validation, IRB review, or regulatory clearance** has been
+  done. This should inform, not replace, professional care.
+
 {PAGE_DISCLAIMER}
 """
 
@@ -497,44 +526,21 @@ with gr.Blocks(title="Skin Disease Detection — Demo") as demo:
             ),
         )
 
-    with gr.Tab("📊 Model Evaluation"):
-        with gr.Accordion("ℹ️ About Model Evaluation", open=False):
-            gr.HTML(data_badge(
-                "🧪 All charts below are computed on <b>held-out test data</b> "
-                "by <code>build_artifacts.py</code> — fully reproducible."
-            ))
-    
-        artifacts = load_performance_artifacts()
-    
-        if not any(artifacts.values()):
-    
+    with gr.Tab("📊 Model Performance"):
+        performance_reports = load_performance_reports()
+        if not performance_reports:
             gr.Markdown(
-                "### No model performance artifacts yet.\n\n"
-                "Run the evaluation notebook and place the generated "
-                "artifacts in the Space's `assets/` folder."
+                "_No model performance data yet._\n\n"
+                "Run either fine-tuning notebook in this repo, then download the "
+                "`*_metrics.json` and `*_confusion_matrix.png` files it produces and place "
+                "them in this Space's `assets/` folder. This tab picks them up automatically "
+                "on the next restart."
             )
-    
         else:
-    
-            gr.Markdown("### Unified Model Evaluation")
-    
-            if artifacts["unified_confusion_matrix"]:
-                gr.Image(
-                    value=artifacts["unified_confusion_matrix"],
-                    label="24-Class Confusion Matrix"
-                )
-    
-            if artifacts["melanoma_threshold_curve"]:
-                gr.Image(
-                    value=artifacts["melanoma_threshold_curve"],
-                    label="Melanoma Precision / Recall vs. Decision Threshold"
-                )
-    
-            if artifacts["cluster_confusion_matrix"]:
-                gr.Image(
-                    value=artifacts["cluster_confusion_matrix"],
-                    label="Disease Similarity Cluster Confusion Matrix"
-                )
+            for report in performance_reports:
+                gr.Markdown(format_performance_markdown(report))
+                if report["image_path"]:
+                    gr.Image(value=report["image_path"], label="Confusion matrix (test set)", show_label=True)
 
     with gr.Tab("ℹ️ About"):
         gr.Markdown(ABOUT_MARKDOWN)
@@ -545,4 +551,4 @@ with gr.Blocks(title="Skin Disease Detection — Demo") as demo:
     )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()