"Vibe-Trading: Your Personal Trading Agent"
Python ai-agentalgorithmic-tradingbacktestingfintechllmmcpmulti-agentpythonquantitative-financetrading
feat(tools): add sentiment analysis tool (#939)
卫斯李6ad0e8d25961
diff --git a/agent/src/tools/sentiment_tool.py b/agent/src/tools/sentiment_tool.pynew file mode 100644index 000000000..c439bc208--- /dev/null+++ b/agent/src/tools/sentiment_tool.py@@ -0,0 +1,146 @@+"""Read-only market sentiment analysis tool.++Two capabilities:+1. ``sentiment_score`` — score arbitrary text [-1, 1] with a lexicon-based scorer+2. ``fear_greed_index`` — fetch the current crypto Fear & Greed Index++All computation is pure Python; the Fear & Greed call is a single HTTP GET+to the free, no-auth alternative.me API.+"""++from __future__ import annotations++import json+import logging+import urllib.request+from typing import Any++from src.agent.tools import BaseTool++logger = logging.getLogger(__name__)++# ── Lexicon (subset, extensible) ─────────────────────────────────────────────+_POSITIVE: frozenset[str] = frozenset({+ "beat", "beats", "bullish", "buy", "cheap", "crush", "crushed",+ "crushing", "gain", "gained", "gaining", "gains", "green",+ "growth", "high", "higher", "hit", "improve", "improved",+ "jump", "jumped", "lead", "leading", "leads", "long", "low",+ "opportunity", "outperform", "outperformed", "profit", "profitable",+ "profits", "raise", "raised", "raises", "rally", "rallied",+ "rebound", "record", "rise", "rises", "rising", "rose",+ "strong", "stronger", "surge", "surged", "surges", "top",+ "undervalued", "up", "upgrade", "upgraded", "upside", "win",+})++_NEGATIVE: frozenset[str] = frozenset({+ "bearish", "bleed", "bottom", "bust", "crash", "crashed",+ "cut", "cuts", "cutting", "decline", "declined", "declining",+ "deficit", "downgrade", "downgraded", "downside", "drop", "dropped",+ "dropping", "fall", "falling", "fell", "fined", "firing",+ "frozen", "headwind", "headwinds", "investigation",+ "lawsuit", "layoff", "layoffs", "litigation", "lose", "loses",+ "losing", "loss", "losses", "low", "lower", "lowered",+ "miss", "missed", "missing", "negative", "overvalued", "penalty",+ "plunge", "plunged", "poor", "probe", "recall", "red",+ "risk", "risks", "sanction", "scandal", "sell", "selling",+ "short", "shrink", "sliding", "slump", "slumped", "turmoil",+ "underperform", "volatile", "volatility", "warn", "warned", "warns", "warning",+ "weak", "weaker", "worry", "worries", "worse", "worst",+})+++def _tokenize(text: str) -> list[str]:+ """Lowercase, split, and strip common punctuation from each token."""+ stripped = [tok.strip(".,;:!?\"'()[]{}") for tok in text.split()]+ return [tok.lower() for tok in stripped if tok and tok.isalpha()]+++def _score_text(text: str) -> dict[str, Any]:+ """Score one text snippet in [-1, 1].++ Returns a dict with ``score`` (float), ``positive`` (int), and ``negative`` (int).+ """+ tokens = _tokenize(text)+ if not tokens:+ return {"score": 0.0, "positive": 0, "negative": 0}+ pos = sum(1 for t in tokens if t in _POSITIVE)+ neg = sum(1 for t in tokens if t in _NEGATIVE)+ hits = pos + neg+ if hits == 0:+ return {"score": 0.0, "positive": 0, "negative": 0}+ return {"score": round((pos - neg) / hits, 4), "positive": pos, "negative": neg}+++def _fetch_fear_greed() -> dict[str, Any] | None:+ """Fetch the latest crypto Fear & Greed Index from alternative.me."""+ url = "https://api.alternative.me/fng/?limit=1"+ try:+ req = urllib.request.Request(url, headers={"User-Agent": "Vibe-Trading/1.0"})+ with urllib.request.urlopen(req, timeout=10) as resp:+ body = json.loads(resp.read().decode())+ except Exception as exc:+ logger.debug("Fear & Greed fetch failed: %s", exc)+ return None+ data = body.get("data", [])+ if not data:+ return None+ item = data[0]+ return {+ "value": int(item.get("value", 0)),+ "classification": item.get("value_classification", "unknown"),+ }+++class SentimentTool(BaseTool):+ """Analyze market sentiment from text and external indices.++ Two modes:+ - ``sentiment_score``: score any free-text (news headline, tweet, etc.) on [-1, 1]+ - ``fear_greed_index``: fetch the crypto Fear & Greed Index (0-100, lower = more fear)+ """++ name = "sentiment"+ description = (+ "Analyze market sentiment. Two modes: (1) 'sentiment_score' scores "+ "arbitrary text on -1 (bearish) to 1 (bullish); (2) 'fear_greed_index' "+ "returns the crypto Fear & Greed Index (0-100, lower = more fear). "+ 'Example: {"mode": "sentiment_score", "text": "Tesla beats earnings estimates"}'+ )+ parameters = {+ "type": "object",+ "properties": {+ "mode": {+ "type": "string",+ "enum": ["sentiment_score", "fear_greed_index"],+ "description": (+ "'sentiment_score' to score text; "+ "'fear_greed_index' for the crypto F&G index."+ ),+ },+ "text": {+ "type": "string",+ "description": "Text to analyze (required when mode='sentiment_score').",+ },+ },+ "required": ["mode"],+ }+ repeatable = True+ is_readonly = True++ def execute(self, **kwargs: Any) -> str:+ mode = str(kwargs.get("mode", "")).strip()++ if mode == "fear_greed_index":+ fg = _fetch_fear_greed()+ if fg is None:+ return json.dumps({"ok": False, "error": "Failed to fetch Fear & Greed Index"})+ return json.dumps({"ok": True, "mode": mode, **fg})++ if mode == "sentiment_score":+ text = str(kwargs.get("text", "")).strip()+ if not text:+ return json.dumps({"ok": False, "error": "text is required for mode 'sentiment_score'"})+ result = _score_text(text)+ return json.dumps({"ok": True, "mode": mode, "text": text[:500], **result})++ return json.dumps({"ok": False, "error": f"Unknown mode: {mode!r}"})diff --git a/agent/tests/test_sentiment_tool.py b/agent/tests/test_sentiment_tool.pynew file mode 100644index 000000000..542913fc6--- /dev/null+++ b/agent/tests/test_sentiment_tool.py@@ -0,0 +1,183 @@+"""Tests for the sentiment analysis tool."""++import json+from unittest.mock import patch++from src.tools.sentiment_tool import (+ SentimentTool,+ _score_text,+ _tokenize,+)+++class TestTokenize:+ def test_empty(self):+ assert _tokenize("") == []++ def test_punctuation_only(self):+ assert _tokenize("!!! $$$") == []++ def test_mixed(self):+ tokens = _tokenize("Tesla beats earnings ESTIMATES!!!")+ assert "tesla" in tokens+ assert "beats" in tokens+ assert "earnings" in tokens+ assert "estimates" in tokens+ assert "!!!" not in " ".join(tokens)+++class TestScoreText:+ def test_strongly_bullish(self):+ r = _score_text("profit surge growth rally record beat upgrade strong")+ assert r["score"] == 1.0+ assert r["positive"] == 8+ assert r["negative"] == 0++ def test_strongly_bearish(self):+ r = _score_text("crash plunge loss decline drop scandal weak downgrade")+ assert r["score"] == -1.0+ assert r["positive"] == 0+ assert r["negative"] == 8++ def test_neutral(self):+ r = _score_text("Tesla announced quarterly results today")+ assert r["score"] == 0.0+ assert r["positive"] == 0+ assert r["negative"] == 0++ def test_flat_is_neutral(self):+ """'flat' was removed from negative terms — financial neutral."""+ r = _score_text("markets flat today")+ assert r["score"] == 0.0++ def test_mixed(self):+ r = _score_text("profit beat expectations but future outlook worry decline")+ # profit, beat = 2 pos; worry, decline = 2 neg; (2-2)/4 = 0+ assert r["score"] == 0.0++ def test_slightly_bullish(self):+ r = _score_text("earnings beat profit growth outlook worry")+ # 4 positive (beat, profit, growth) vs 1 negative (worry) → (3-1)/4 = 0.5+ # Actually: beat, profit, growth = 3 pos; worry = 1 neg; score = (3-1)/4 = 0.5+ assert r["score"] == 0.5++ def test_real_headlines(self):+ """Verify scoring makes sense on realistic financial headlines."""+ assert _score_text("Tesla crushes earnings estimates, stock surges")["score"] > 0.5+ assert _score_text("Company warns of revenue miss, shares plunge")["score"] < -0.5+ assert _score_text("Fed holds rates steady as expected")["score"] == 0.0++ def test_empty_text(self):+ r = _score_text("")+ assert r["score"] == 0.0+ assert r["positive"] == 0++ def test_no_alpha_tokens(self):+ r = _score_text("123 456 !!! ???")+ assert r["score"] == 0.0+++class TestSentimentTool:+ def test_missing_mode(self):+ tool = SentimentTool()+ result = json.loads(tool.execute())+ assert result["ok"] is False+ assert "Unknown mode" in result["error"]++ def test_unknown_mode(self):+ tool = SentimentTool()+ result = json.loads(tool.execute(mode="invalid"))+ assert result["ok"] is False++ def test_sentiment_score_missing_text(self):+ tool = SentimentTool()+ result = json.loads(tool.execute(mode="sentiment_score"))+ assert result["ok"] is False+ assert "text" in result["error"]++ def test_sentiment_score_success(self):+ tool = SentimentTool()+ result = json.loads(tool.execute(mode="sentiment_score", text="profit surge growth"))+ assert result["ok"] is True+ assert result["score"] == 1.0++ def test_sentiment_text_truncated(self):+ tool = SentimentTool()+ long_text = "bullish " * 1000+ result = json.loads(tool.execute(mode="sentiment_score", text=long_text))+ assert result["ok"] is True+ assert len(result["text"]) <= 500++ def test_fear_greed_success(self):+ tool = SentimentTool()+ mock_data = json.dumps({+ "data": [{"value": "28", "value_classification": "Fear"}]+ }).encode()+ with patch("urllib.request.urlopen", return_value=type("m", (), {"read": lambda s: mock_data, "__enter__": lambda s: s, "__exit__": lambda s,*a: None})()):+ result = json.loads(tool.execute(mode="fear_greed_index"))+ assert result["ok"] is True+ assert result["value"] == 28+ assert result["classification"] == "Fear"++ def test_fear_greed_failure(self):+ tool = SentimentTool()+ with patch("urllib.request.urlopen", side_effect=OSError("network down")):+ result = json.loads(tool.execute(mode="fear_greed_index"))+ assert result["ok"] is False+ assert "Failed to fetch" in result["error"]++ def test_fear_greed_empty_data(self):+ tool = SentimentTool()+ mock_data = json.dumps({"data": []}).encode()+ with patch("urllib.request.urlopen", return_value=type("m", (), {"read": lambda s: mock_data, "__enter__": lambda s: s, "__exit__": lambda s,*a: None})()):+ result = json.loads(tool.execute(mode="fear_greed_index"))+ assert result["ok"] is False++ def test_fear_greed_malformed_json(self):+ tool = SentimentTool()+ mock_data = b"not json"+ with patch("urllib.request.urlopen", return_value=type("m", (), {"read": lambda s: mock_data, "__enter__": lambda s: s, "__exit__": lambda s,*a: None})()):+ result = json.loads(tool.execute(mode="fear_greed_index"))+ assert result["ok"] is False++ def test_fear_greed_missing_value(self):+ tool = SentimentTool()+ mock_data = json.dumps({"data": [{"value_classification": "Neutral"}]}).encode()+ with patch("urllib.request.urlopen", return_value=type("m", (), {"read": lambda s: mock_data, "__enter__": lambda s: s, "__exit__": lambda s,*a: None})()):+ result = json.loads(tool.execute(mode="fear_greed_index"))+ assert result["ok"] is True+ assert result["value"] == 0 # default int++ def test_sentiment_unicode(self):+ """Non-ASCII text should not crash."""+ tool = SentimentTool()+ result = json.loads(tool.execute(mode="sentiment_score", text="特斯拉 profit 增长 surge 🚀"))+ assert result["ok"] is True+ assert result["score"] == 1.0 # profit + surge++ def test_sentiment_very_long(self):+ """Very long text should not crash or timeout."""+ tool = SentimentTool()+ result = json.loads(tool.execute(mode="sentiment_score", text="profit " * 5000))+ assert result["ok"] is True+ # 5000 "profit" tokens → all positive → score = 1.0+ assert result["score"] == 1.0++ def test_sentiment_no_alpha(self):+ tool = SentimentTool()+ result = json.loads(tool.execute(mode="sentiment_score", text="12345 67890 !@#$%"))+ assert result["ok"] is True+ assert result["score"] == 0.0++ def test_non_string_text_coerced(self):+ """Non-string text should be coerced to string, not crash."""+ tool = SentimentTool()+ result = json.loads(tool.execute(mode="sentiment_score", text=12345))+ assert result["ok"] is True+ assert isinstance(result["text"], str)++ def test_non_string_mode_coerced(self):+ """Non-string mode should be coerced."""+ tool = SentimentTool()+ result = json.loads(tool.execute(mode=999))+ assert result["ok"] is False