Update app.py

app.py

@@ -1,61 +1,226 @@
+"""
+Bitcoin Market Signal Engine — Hugging Face Spaces
+====================================================
+Fetches Bitcoin headlines from Google News RSS, runs DistilBERT sentiment
+analysis, computes a confidence-weighted market signal, and returns JSON.
+
+Requirements (see requirements.txt):
+    transformers, torch, gradio, feedparser, requests
+"""
+
+import re
+import json
 import feedparser
+import gradio as gr
+from datetime import datetime, timezone
 from transformers import pipeline
-import re
 
-# DistilBERT modelini yükle (CPU üzerinde)
-model = pipeline("sentiment-analysis", model="distilbert-base-uncased", device=-1)
-
-# Bitcoin haberleri için RSS feed URL'si
-rss_url = "https://news.google.com/rss/search?q=Bitcoin&hl=en-US&gl=US&ceid=US:en"
-
-# RSS feed'i al
-feed = feedparser.parse(rss_url)
-
-# Toplam skor ve haber sayısı
-total_score = 0
-count = 0
-
-# Ön işleme fonksiyonu
-def preprocess_title(title):
-    # % -> percent, $ -> USD
-    title = title.replace("%", " percent").replace("$", " USD")
-    # Gereksiz sembolleri temizle
-    title = re.sub(r"[^a-zA-Z0-9\s]", "", title)
-    return title
-
-# Her başlığı analiz et
-for entry in feed.entries:
-    raw_title = entry.title
-    title = preprocess_title(raw_title)
-    
-    sentiment = model(title)[0]
-
-    label = sentiment['label'].upper()
-    score = sentiment['score']  # modelin güven skorunu kullan
-
-    # Ağırlıklı puanlama
-    if label == "POSITIVE":
-        total_score += score
-    elif label == "NEGATIVE":
-        total_score -= score
-    else:  # NEUTRAL
-        total_score += 0
-
-    count += 1
-
-# Ortalama skor
-average_score = total_score / count if count > 0 else 0
-
-# Long / Short / Neutral kararı
-threshold = 0.05  # küçük dalgalanmaları görmezden gel
-if average_score > threshold:
-    signal = "LONG 🟢"
-elif average_score < -threshold:
-    signal = "SHORT 🔴"
-else:
-    signal = "NEUTRAL ⚪️"
-
-# Sonuçları yazdır
-print(f"Toplam Başlık Sayısı: {count}")
-print(f"Ortalama Duygu Skoru: {average_score:.2f}")
-print(f"Tahmini Piyasa Sinyali: {signal}")
+# ── Model (loaded once at startup) ──────────────────────────────────────────
+print("Loading DistilBERT sentiment model...")
+sentiment_pipe = pipeline(
+    "sentiment-analysis",
+    model="distilbert-base-uncased-finetuned-sst-2-english",
+    device=-1,          # CPU; set to 0 if GPU available
+    truncation=True,
+    max_length=512,
+)
+print("Model ready.")
+
+RSS_URL = (
+    "https://news.google.com/rss/search"
+    "?q=Bitcoin&hl=en-US&gl=US&ceid=US:en"
+)
+
+
+# ── Helpers ──────────────────────────────────────────────────────────────────
+
+def preprocess(text: str) -> str:
+    """Normalise a headline for the model."""
+    text = text.replace("%", "percent").replace("$", "USD")
+    text = re.sub(r"[^a-zA-Z0-9 ]", " ", text)
+    text = re.sub(r"\s+", " ", text).strip().lower()
+    return text
+
+
+def fetch_headlines(max_items: int = 30) -> list[dict]:
+    """Pull headlines from Google News RSS via feedparser."""
+    feed = feedparser.parse(RSS_URL)
+    entries = feed.entries[:max_items]
+    if not entries:
+        raise ValueError("RSS feed returned no items. Try again shortly.")
+    return [
+        {
+            "raw": e.get("title", ""),
+            "link": e.get("link", ""),
+            "published": e.get("published", ""),
+        }
+        for e in entries
+    ]
+
+
+def classify(headlines: list[dict]) -> list[dict]:
+    """
+    Run DistilBERT on preprocessed headlines.
+    Maps POSITIVE → +score, NEGATIVE → -score, NEUTRAL → 0.
+    DistilBERT SST-2 only emits POSITIVE/NEGATIVE; we keep that faithful.
+    """
+    cleaned = [preprocess(h["raw"]) for h in headlines]
+    raw_preds = sentiment_pipe(cleaned, batch_size=16)
+
+    results = []
+    for h, pred in zip(headlines, raw_preds):
+        label = pred["label"]          # "POSITIVE" or "NEGATIVE"
+        score = float(pred["score"])   # confidence ∈ (0.5, 1.0]
+        numeric = score if label == "POSITIVE" else -score
+        results.append(
+            {
+                **h,
+                "clean": preprocess(h["raw"]),
+                "label": label,
+                "confidence": round(score, 6),
+                "numeric_score": round(numeric, 6),
+            }
+        )
+    return results
+
+
+def compute_signal(scored: list[dict]) -> dict:
+    """Weighted average → LONG / SHORT / NEUTRAL signal."""
+    avg = sum(s["numeric_score"] for s in scored) / len(scored)
+    signal = "LONG" if avg > 0.05 else "SHORT" if avg < -0.05 else "NEUTRAL"
+
+    by_score = sorted(scored, key=lambda x: x["numeric_score"], reverse=True)
+    top_pos = [s["raw"] for s in by_score if s["numeric_score"] > 0][:5]
+    top_neg = [s["raw"] for s in reversed(by_score) if s["numeric_score"] < 0][:5]
+
+    label_counts = {"POSITIVE": 0, "NEGATIVE": 0}
+    for s in scored:
+        label_counts[s["label"]] = label_counts.get(s["label"], 0) + 1
+
+    return {
+        "timestamp": datetime.now(timezone.utc).isoformat(),
+        "headline_count": len(scored),
+        "label_distribution": label_counts,
+        "average_score": round(avg, 6),
+        "market_signal": signal,
+        "top_positive_headlines": top_pos,
+        "top_negative_headlines": top_neg,
+    }
+
+
+# ── Gradio handler ───────────────────────────────────────────────────────────
+
+def run_analysis(max_headlines: int, show_detail: bool):
+    """Main Gradio callback."""
+    try:
+        # 1. Fetch
+        headlines = fetch_headlines(int(max_headlines))
+
+        # 2. Classify
+        scored = classify(headlines)
+
+        # 3. Signal
+        result = compute_signal(scored)
+
+        # 4. Build outputs
+        signal_md = (
+            f"## Signal: **{result['market_signal']}**\n\n"
+            f"- **Average score:** `{result['average_score']}`\n"
+            f"- **Headlines analysed:** {result['headline_count']}\n"
+            f"- **Positive:** {result['label_distribution'].get('POSITIVE', 0)}  "
+            f"**Negative:** {result['label_distribution'].get('NEGATIVE', 0)}\n"
+            f"- **Timestamp:** {result['timestamp']}"
+        )
+
+        # Core JSON (spec-compliant)
+        core_json = {
+            k: result[k]
+            for k in (
+                "timestamp",
+                "headline_count",
+                "average_score",
+                "market_signal",
+                "top_positive_headlines",
+                "top_negative_headlines",
+            )
+        }
+        json_out = json.dumps(core_json, indent=2)
+
+        # Detail table
+        if show_detail:
+            detail_rows = [
+                [
+                    i + 1,
+                    s["label"],
+                    f"{s['numeric_score']:+.4f}",
+                    s["raw"][:120],
+                ]
+                for i, s in enumerate(scored)
+            ]
+        else:
+            detail_rows = []
+
+        return signal_md, json_out, detail_rows, ""
+
+    except Exception as exc:
+        return "", "", [], f"❌ Error: {exc}"
+
+
+# ── UI ───────────────────────────────────────────────────────────────────────
+
+with gr.Blocks(
+    title="Bitcoin Market Signal Engine",
+    theme=gr.themes.Monochrome(),
+    css="""
+        #signal-md { font-size: 1.15rem; }
+        #json-box textarea { font-family: monospace; font-size: 0.82rem; }
+        footer { display: none !important; }
+    """,
+) as demo:
+
+    gr.Markdown(
+        """
+        # ₿ Bitcoin Market Signal Engine
+        Fetches live Bitcoin headlines → DistilBERT sentiment → confidence-weighted signal.
+        """
+    )
+
+    with gr.Row():
+        max_h = gr.Slider(5, 30, value=20, step=1, label="Max headlines to fetch")
+        show_d = gr.Checkbox(value=True, label="Show per-headline detail table")
+        run_btn = gr.Button("▶ Run Analysis", variant="primary")
+
+    with gr.Row():
+        signal_out = gr.Markdown(elem_id="signal-md", label="Market Signal")
+
+    json_out = gr.Code(
+        label="JSON Output (spec-compliant)",
+        language="json",
+        elem_id="json-box",
+    )
+
+    detail_tbl = gr.Dataframe(
+        headers=["#", "Label", "Score", "Headline"],
+        label="Headline Sentiment Detail",
+        wrap=True,
+    )
+
+    error_box = gr.Textbox(label="Errors", visible=True, interactive=False)
+
+    run_btn.click(
+        fn=run_analysis,
+        inputs=[max_h, show_d],
+        outputs=[signal_out, json_out, detail_tbl, error_box],
+    )
+
+    gr.Markdown(
+        """
+        ---
+        **Pipeline:** RSS fetch → preprocess (% → percent, $ → USD, strip special chars, lowercase)
+        → `distilbert-base-uncased-finetuned-sst-2-english` → confidence-weighted average
+        → LONG (avg > 0.05) / SHORT (avg < -0.05) / NEUTRAL
+        """
+    )
+
+if __name__ == "__main__":
+    demo.launch()