ghstedpixel/BRIE
0
1import os2import pickle3import subprocess4import sqlite35from datetime import datetime6 7import duckdb8import gradio as gr9import numpy as np10import pandas as pd11import pandas_ta as ta12import plotly.graph_objects as go13import yfinance as yf14from xgboost import XGBRegressor15from huggingface_hub import HfApi, hf_hub_download16from dotenv import load_dotenv17 18# --- HEALTH AGENT BLOCK ---19os.environ["PYTHONUNBUFFERED"] = "1"20 21def run_agent():22 print("🩺 Starting Health Agent...")23 result = subprocess.run(24 ["python", "health_agent.py"],25 cwd=".",26 capture_output=False,27 text=True28 )29 print(f"🩺 Agent complete: {result.returncode}")30 31run_agent()32print("🩺 Health Agent launched & complete!")33 34load_dotenv()35 36# --- UNIFIED PATHS & CONSTANTS ---37WATCHLIST_DB = os.path.join(os.getcwd(), "watchlist_manager.db")38DB_PATH = "/tmp/brie_brain.db"39MODEL_PATH = "/tmp/brie_model.pkl"40 41MOMENTUM_THRESHOLD = 7542OVERBOUGHT_RSI = 7043 44FEATURES = [45 "rsi", "macd_h", "ema_20_dist", "ema_50_dist", "vwap_dist",46 "atr_pct", "bb_width", "vol_spike", "ret_1d", "ret_5d",47 "day_of_week", "sentiment_score", "gap_pct", "vix_level"48]49 50SECTOR_ETFS = ["XLF", "XLK", "XLE", "XLY", "XLP", "XLV", "XLI", "XLU", "XLB", "XLRE", "XLC"]51VOLATILITY_INDEX = "^VIX"52 53HF_TOKEN = os.getenv("HF_TOKEN")54HF_REPO_ID = os.getenv("HF_REPO_ID")55 56# --- DATABASE HELPERS ---57def get_current_watchlist():58 conn = sqlite3.connect(WATCHLIST_DB)59 cursor = conn.cursor()60 cursor.execute("CREATE TABLE IF NOT EXISTS watchlist (ticker TEXT PRIMARY KEY)")61 cursor.execute("SELECT ticker FROM watchlist ORDER BY ticker")62 data = cursor.fetchall()63 conn.close()64 return pd.DataFrame(data, columns=["Ticker"])65 66def add_ticker_to_db(ticker):67 if ticker:68 ticker = ticker.upper().strip()69 conn = sqlite3.connect(WATCHLIST_DB)70 cursor = conn.cursor()71 cursor.execute("INSERT OR IGNORE INTO watchlist (ticker) VALUES (?)", (ticker,))72 conn.commit()73 conn.close()74 return get_current_watchlist(), ""75 76def remove_ticker_from_db(ticker):77 if ticker:78 ticker = ticker.upper().strip()79 conn = sqlite3.connect(WATCHLIST_DB)80 cursor = conn.cursor()81 cursor.execute("DELETE FROM watchlist WHERE ticker = ?", (ticker,))82 conn.commit()83 conn.close()84 return get_current_watchlist(), ""85 86def trigger_discord():87 try:88 from brie_agent import run_alpaca_briefing89 watchlist_df = get_current_watchlist()90 current_list = watchlist_df["Ticker"].tolist()91 if not current_list:92 return "⚠️ Watchlist is empty."93 return str(run_alpaca_briefing(current_list))94 except Exception as e:95 return f"❌ Bridge Error: {str(e)}"96 97# --- CLOUD SYNC & DB INIT ---98def init_db():99 with duckdb.connect(DB_PATH) as con:100 feat_cols = ", ".join([f'"{f}" FLOAT' for f in FEATURES])101 con.execute(102 f"""103 CREATE TABLE IF NOT EXISTS feature_store (104 ticker TEXT,105 timestamp TIMESTAMP,106 {feat_cols},107 target_next_day_ret FLOAT,108 PRIMARY KEY (ticker, timestamp)109 )110 """111 )112 con.execute(113 """114 CREATE TABLE IF NOT EXISTS performance_logs (115 ticker TEXT,116 timestamp TIMESTAMP,117 predicted_ret FLOAT,118 entry_price FLOAT,119 actual_ret FLOAT,120 is_verified BOOLEAN DEFAULT FALSE121 )122 """123 )124 125def load_from_cloud():126 if not HF_TOKEN or not HF_REPO_ID:127 return128 try:129 for f in ["brie_model.pkl", "brie_brain.db"]:130 hf_hub_download(131 repo_id=HF_REPO_ID,132 filename=f,133 local_dir="/tmp",134 repo_type="dataset",135 token=HF_TOKEN136 )137 print("✅ Cloud models/data loaded")138 except Exception as e:139 print(f"⚠️ Cloud load failed: {e}")140 141def save_to_cloud():142 if not HF_TOKEN or not HF_REPO_ID:143 return "⚠️ HF_TOKEN missing"144 api = HfApi()145 try:146 for path, filename in [(MODEL_PATH, "brie_model.pkl"), (DB_PATH, "brie_brain.db")]:147 if os.path.exists(path):148 api.upload_file(149 path_or_fileobj=path,150 path_in_repo=filename,151 repo_id=HF_REPO_ID,152 repo_type="dataset",153 token=HF_TOKEN154 )155 return "✅ Sync Success"156 except Exception as e:157 return f"❌ Sync Failed: {str(e)[:30]}"158 159# --- ANALYTICS ENGINE ---160def fetch_history_yf(ticker, period="5y", interval="1d"):161 try:162 df = yf.download(163 ticker,164 period=period,165 interval=interval,166 auto_adjust=False,167 progress=False168 )169 if df.empty:170 return pd.DataFrame()171 if isinstance(df.columns, pd.MultiIndex):172 df.columns = df.columns.get_level_values(0)173 df.columns = [str(c).lower().replace(" ", "_") for c in df.columns]174 df.index = pd.to_datetime(df.index).tz_localize(None)175 return df[["open", "high", "low", "close", "volume"]]176 except:177 return pd.DataFrame()178 179def build_feature_set(df, sentiment_val=0.0, labeling=False, vix_df=None):180 if len(df) < 100:181 return pd.DataFrame()182 ind = pd.DataFrame(index=df.index)183 ind["rsi"] = ta.rsi(df["close"], length=14).fillna(50.0)184 macd = ta.macd(df["close"])185 ind["macd_h"] = macd.iloc[:, 1].fillna(0.0) if macd is not None else 0.0186 bbands = ta.bbands(df["close"], length=20, std=2)187 ind["bb_width"] = ((bbands.iloc[:, 2] - bbands.iloc[:, 0]) / bbands.iloc[:, 1]).fillna(0.0) if bbands is not None else 0.0188 for length in [20, 50]:189 ema = ta.ema(df["close"], length=length)190 ind[f"ema_{length}_dist"] = ((df["close"] - ema) / ema).fillna(0.0)191 vwap = ta.vwap(df["high"], df["low"], df["close"], df["volume"])192 ind["vwap_dist"] = ((df["close"] - vwap) / vwap).fillna(0.0)193 ind["atr_pct"] = (ta.atr(df["high"], df["low"], df["close"]) / df["close"]).fillna(0.0)194 ind["vol_spike"] = (df["volume"] / df["volume"].rolling(20).mean()).fillna(1.0)195 ind["ret_1d"] = df["close"].pct_change(1).fillna(0.0)196 ind["ret_5d"] = df["close"].pct_change(5).fillna(0.0)197 ind["day_of_week"] = df.index.dayofweek.astype(float)198 ind["sentiment_score"] = float(sentiment_val)199 ind["gap_pct"] = ((df["open"] - df["close"].shift(1)) / df["close"].shift(1)).fillna(0.0)200 if vix_df is not None and not vix_df.empty:201 ind["vix_level"] = vix_df["close"].reindex(df.index, method="ffill").fillna(20.0)202 else:203 ind["vix_level"] = pd.Series(20.0, index=df.index)204 if labeling:205 ind["target_next_day_ret"] = df["close"].pct_change(1).shift(-1)206 cols = FEATURES + (["target_next_day_ret"] if labeling else [])207 return ind[cols].dropna()208 209def calculate_momentum_score(pred_return, rsi, vol_spike):210 normalized_pred = np.clip((pred_return + 5) * 10, 0, 100)211 vol_bonus = 15 if vol_spike > 1.2 else 0212 rsi_penalty = (rsi - OVERBOUGHT_RSI) * 2 if rsi > OVERBOUGHT_RSI else 0213 return np.clip((normalized_pred * 0.7) + vol_bonus - rsi_penalty, 0, 100)214 215def log_prediction_to_db(ticker, pred_val, entry_price):216 init_db()217 with duckdb.connect(DB_PATH) as con:218 con.execute(219 "INSERT INTO performance_logs (ticker, timestamp, predicted_ret, entry_price) VALUES (?, ?, ?, ?)",220 (ticker, datetime.now(), pred_val, entry_price)221 )222 223def update_performance_ledger():224 init_db()225 with duckdb.connect(DB_PATH) as con:226 pending = con.execute(227 "SELECT * FROM performance_logs WHERE is_verified = FALSE AND timestamp < (CURRENT_TIMESTAMP - INTERVAL '1 DAY')"228 ).df()229 for _, row in pending.iterrows():230 current_bars = fetch_history_yf(row["ticker"], period="1d")231 if not current_bars.empty:232 actual_ret = ((current_bars["close"].iloc[-1] - row["entry_price"]) / row["entry_price"]) * 100233 con.execute(234 "UPDATE performance_logs SET actual_ret = ?, is_verified = TRUE WHERE ticker = ? AND timestamp = ?",235 (actual_ret, row["ticker"], row["timestamp"])236 )237 return "✅ Ledger Updated"238 239def get_performance_dashboard():240 update_performance_ledger()241 init_db()242 with duckdb.connect(DB_PATH) as con:243 df = con.execute("SELECT * FROM performance_logs WHERE is_verified = TRUE").df()244 if df.empty:245 return "### ⚠️ No verified data yet. (Wait 24h)", None246 df["error"] = df["predicted_ret"] - df["actual_ret"]247 win_rate = (np.sign(df["predicted_ret"]) == np.sign(df["actual_ret"])).mean() * 100248 fig = go.Figure()249 fig.add_trace(go.Scatter(x=df["predicted_ret"], y=df["actual_ret"], mode="markers", name="Predictions", marker=dict(color="#FFC107")))250 fig.update_layout(template="plotly_dark", title="Prediction vs. Actual", xaxis_title="AI Forecast %", yaxis_title="Actual Result %")251 stats = f"### Directional Win Rate: {win_rate:.1f}% | N={len(df)}"252 return stats, fig253 254def run_incremental_training(batch_str):255 tickers = [t.strip().upper() for t in batch_str.split(",") if t.strip()]256 vix_data = fetch_history_yf(VOLATILITY_INDEX)257 yield "📥 Syncing Data..."258 all_data = []259 for t in tickers:260 bars = fetch_history_yf(t)261 if bars.empty:262 continue263 f_set = build_feature_set(bars, vix_df=vix_data, labeling=True)264 if f_set.empty:265 continue266 f_set["ticker"] = t267 f_set = f_set.reset_index().rename(columns={"Date": "timestamp", "index": "timestamp"})268 all_data.append(f_set)269 if not all_data:270 yield "❌ No data found."271 return272 df = pd.concat(all_data, ignore_index=True)273 init_db()274 with duckdb.connect(DB_PATH) as con:275 con.register("df_tmp", df)276 con.execute("INSERT OR IGNORE INTO feature_store SELECT ticker, timestamp, " + ", ".join([f'"{f}"' for f in FEATURES]) + ", target_next_day_ret FROM df_tmp")277 data = con.execute("SELECT * FROM feature_store").df()278 yield "🧠 Training XGBoost..."279 model = XGBRegressor(n_estimators=200, max_depth=7, learning_rate=0.02)280 model.fit(data[FEATURES], data["target_next_day_ret"])281 with open(MODEL_PATH, "wb") as f:282 pickle.dump(model, f)283 yield f"✅ Refreshed! {save_to_cloud()}"284 285def analyze_asset(ticker):286 ticker = ticker.upper().strip()287 bars = fetch_history_yf(ticker)288 if bars.empty:289 return "❌ No data", None, "", "", ""290 last_price = bars["close"].iloc[-1]291 pred_text = "⚠️ Model not trained"292 if os.path.exists(MODEL_PATH):293 with open(MODEL_PATH, "rb") as f:294 reg = pickle.load(f)295 feat = build_feature_set(bars, vix_df=fetch_history_yf(VOLATILITY_INDEX, period="1mo"))296 if not feat.empty:297 latest = feat.tail(1)298 pred_val = float(reg.predict(latest[FEATURES])[0] * 100)299 log_prediction_to_db(ticker, pred_val, last_price)300 m_score = calculate_momentum_score(pred_val, float(latest["rsi"].iloc[0]), float(latest["vol_spike"].iloc[0]))301 if m_score > MOMENTUM_THRESHOLD:302 signal, color = "🔥 STRONG", "#00FF00"303 elif m_score > 50:304 signal, color = "⚖️ TREND", "#FFC107"305 else:306 signal, color = "⚠️ WEAK", "#FF5252"307 pred_text = f"<div style='text-align: center; border: 1px solid {color}; padding: 10px;'><h2>{m_score:.1f}/100</h2><h3>{signal}</h3><p>Forecast: {pred_val:+.2f}%</p></div>"308 fig = go.Figure(data=[go.Candlestick(x=bars.index, open=bars["open"], high=bars["high"], low=bars["low"], close=bars["close"])])309 fig.update_layout(template="plotly_dark", xaxis_rangeslider_visible=False, height=450)310 return f"### {ticker} Analysis", fig, pred_text, f"${last_price:.2f}", ""311 312def get_sector_analysis():313 if not os.path.exists(MODEL_PATH):314 return "❌ Train First", None, ""315 with open(MODEL_PATH, "rb") as f:316 reg = pickle.load(f)317 vix = fetch_history_yf(VOLATILITY_INDEX)318 rows = []319 for t in SECTOR_ETFS:320 bars = fetch_history_yf(t, period="2y")321 feat = build_feature_set(bars, vix_df=vix)322 if feat.empty:323 continue324 rows.append({"Sector": t, "Return": reg.predict(feat.tail(1)[FEATURES])[0] * 100})325 if not rows:326 return "❌ No sector data", None, ""327 df_res = pd.DataFrame(rows).sort_values("Return", ascending=False)328 fig = go.Figure(go.Bar(x=df_res["Sector"], y=df_res["Return"], marker_color="#FFC107"))329 fig.update_layout(template="plotly_dark", title="Sector Sentiment", height=400)330 return f"### Strongest: {df_res['Sector'].iloc[0]}", fig, "Macro synced."331 332# --- GRADIO INTERFACE ---333with gr.Blocks() as brie:334 gr.Markdown("# 🧀 Brie Ultra AI")335 with gr.Tabs():336 with gr.Tab("🔍 Scanner"):337 with gr.Row():338 ticker_in = gr.Textbox(label="Ticker", scale=4)339 scan_btn = gr.Button("Predict", variant="primary", scale=1)340 chart = gr.Plot()341 with gr.Row():342 res_txt, pred_out, price_lbl = gr.Markdown(), gr.HTML(), gr.Label(label="Last Close")343 scan_btn.click(analyze_asset, ticker_in, [res_txt, chart, pred_out, price_lbl])344 with gr.Tab("🎯 Accuracy Dashboard"):345 perf_btn = gr.Button("Refresh Accuracy Stats")346 perf_stats, perf_chart = gr.Markdown(), gr.Plot()347 perf_btn.click(get_performance_dashboard, None, [perf_stats, perf_chart])348 with gr.Tab("📊 Sector Impact"):349 sector_btn = gr.Button("Analyze Macro")350 sector_top, sector_chart, sector_ai = gr.Markdown(), gr.Plot(), gr.Markdown()351 sector_btn.click(get_sector_analysis, None, [sector_top, sector_chart, sector_ai])352 with gr.Tab("⚙️ Continuous Learning"):353 gr.Markdown("### Watchlist")354 with gr.Row():355 ticker_input = gr.Textbox(label="Ticker", scale=3)356 add_btn, remove_btn = gr.Button("➕ Add"), gr.Button("🗑️ Remove")357 watchlist_display = gr.Dataframe(headers=["Ticker"], value=get_current_watchlist())358 gr.Markdown("---")359 cal_btn = gr.Button("Learn & Sync", variant="stop")360 cal_out = gr.Markdown("Ready.")361 cal_btn.click(run_incremental_training, gr.Textbox(value=",".join(SECTOR_ETFS), visible=False), [cal_out])362 gr.Markdown("---")363 discord_btn = gr.Button("🚀 Send Briefing", variant="primary")364 discord_status = gr.Label(value="Ready")365 add_btn.click(add_ticker_to_db, [ticker_input], [watchlist_display, ticker_input])366 remove_btn.click(remove_ticker_from_db, [ticker_input], [watchlist_display, ticker_input])367 discord_btn.click(trigger_discord, None, [discord_status])368 369if __name__ == "__main__":370 init_db()371 load_from_cloud()372 init_db()373 brie.queue().launch(374 theme=gr.themes.Default(primary_hue="amber"),375 server_name="0.0.0.0",376 server_port=int(os.getenv("PORT", 7860))377 )