""" app.py — Sniper Model Evaluator Space Two modes: Backtester and Model Evaluator. """ import sys import json import time import logging from datetime import datetime, date from pathlib import Path import numpy as np import pandas as pd import gradio as gr sys.path.insert(0, str(Path(__file__).parent)) # Must run before any joblib.load — injects SniperModel into __main__ # so pickles trained with __main__.SniperModel deserialize correctly. from src.sniper_model import patch_main as _patch_main _patch_main() from src.registry import ( discover_repos, load_bundle, get_all_bundle_labels, find_bundle_by_label, ModelRepo, ArtifactBundle, ) from src.data_loader import ( download_ticker_batch, filter_ticker_data, extract_market_series, ) from src.backtester import run_backtest, BacktestConfig, PRESETS from src.evaluator import run_evaluation, DIMENSION_WEIGHTS from src.charts import ( nav_chart, monthly_returns_heatmap, exit_reasons_chart, trade_return_distribution, radar_chart, reliability_diagram, regime_heatmap, feature_psi_chart, multi_model_comparison, ) logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] %(message)s") logger = logging.getLogger("SniperApp") # --------------------------------------------------------------------------- # Global state # --------------------------------------------------------------------------- _repos: list[ModelRepo] = [] _loaded_bundles: dict[str, ArtifactBundle] = {} # label -> loaded bundle _eval_results: list = [] # for multi-model comparison _eval_labels: list[str] = [] TICKER_UNIVERSE_PRESETS = { "All (~800 tickers)": None, # None = use bundle's full universe "S&P 500 subset (200)": [ # representative 200-ticker subset "AAPL","MSFT","AMZN","GOOGL","META","NVDA","TSLA","BRK-B","UNH","JNJ", "XOM","JPM","V","PG","MA","CVX","HD","LLY","ABBV","MRK","PEP","KO", "BAC","PFE","COST","TMO","AVGO","CSCO","ABT","ACN","WMT","CRM","NEE", "TXN","DHR","VZ","AMGN","NKE","MDT","UPS","LIN","PM","RTX","QCOM", "HON","T","SBUX","INTU","AMAT","ISRG","GS","MS","SCHW","AXP","BLK", "C","USB","TGT","GILD","CVS","ZTS","SYK","ELV","ADP","NOW","SPGI", "CL","MO","DUK","SO","D","EXC","AEP","SRE","PCG","WEC","ES","XEL", "LMT","NOC","GD","BA","RTX","HII","TDG","HWM","HEI","LDOS", "CAT","DE","EMR","ETN","ITW","PH","ROK","DOV","CARR","OTIS", "F","GM","APTV","BWA","LEA","MGA","AZO","ORLY","GPC","AAP", "AMZN","EBAY","ETSY","ROST","TJX","LOW","HD","WST","SHW","RPM", "ECL","PPG","EMN","LYB","APD","CF","MOS","NUE","FCX","FMC", "BIIB","REGN","VRTX","ILMN","IDXX","MRNA","HOLX","BIO","ABMD", "MCO","ICE","CME","CBOE","NDAQ","MSCI","FDS","MKTX","BR","AMP", "TROW","BEN","IVZ","FHI","AMG","SEIC","APAM","VRTS","HLNE","PIPR", "PLD","EQIX","AMT","CCI","SBAC","ARE","AVB","EQR","MAA","UDR", "WM","RSG","CWST","CLH","GFL","WCN","SRCL","ECOL","ARIS","US", "FDX","UPS","XPO","ODFL","SAIA","WERN","JBHT","ARCB","CHRW","EXPD", ], "Custom (paste tickers)": "custom", } GRADE_COLORS = { "A+": "#00d4aa", "A": "#00d4aa", "A-": "#34d399", "B+": "#60a5fa", "B": "#60a5fa", "B-": "#93c5fd", "C+": "#f5a623", "C": "#f5a623", "C-": "#fbbf24", "D+": "#ff4d6a", "D": "#ff4d6a", "F": "#dc2626", } # --------------------------------------------------------------------------- # Shared helpers # --------------------------------------------------------------------------- def _refresh_registry(progress=gr.Progress()): global _repos messages = [] def _cb(msg): messages.append(msg) progress(0, desc=msg) progress(0, desc="Scanning HuggingFace for Sniper repositories...") _repos = discover_repos(progress_cb=_cb) labels = get_all_bundle_labels(_repos) if not labels: return ( gr.Dropdown(choices=[], value=None, label="Model / Run"), gr.Dropdown(choices=[], value=None, label="Model / Run (Evaluator)"), "⚠️ No models found. Check that Arkm20/sniper-model-* repos are public.", ) status = f"✅ Found {len(_repos)} repo(s), {len(labels)} run(s) total.\n" + "\n".join(f" · {l}" for l in labels) return ( gr.Dropdown(choices=labels, value=labels[0], label="Model / Run"), gr.Dropdown(choices=labels, value=labels[0], label="Model / Run (Evaluator)"), status, ) def _get_or_load_bundle(label: str, progress_cb=None) -> ArtifactBundle | None: global _loaded_bundles if label in _loaded_bundles: return _loaded_bundles[label] bundle = find_bundle_by_label(_repos, label) if bundle is None: return None bundle = load_bundle(bundle, progress_cb=progress_cb) _loaded_bundles[label] = bundle return bundle def _universe_for_preset(preset_name: str, custom_text: str, bundle) -> list[str]: if preset_name == "Custom (paste tickers)": tickers = [t.strip().upper() for t in custom_text.replace(",", " ").split() if t.strip()] return tickers if tickers else (bundle.metadata.get("configuration", {}).get("TICKER_UNIVERSE", []) or []) if preset_name == "S&P 500 subset (200)": return TICKER_UNIVERSE_PRESETS["S&P 500 subset (200)"] # "All" - use training universe from bundle metadata if available, else hardcoded return bundle.metadata.get("configuration", {}).get("TICKER_UNIVERSE", []) or [] # --------------------------------------------------------------------------- # Backtester runner # --------------------------------------------------------------------------- def run_backtester( model_label, bt_start, bt_end, initial_cash, conviction_threshold, max_positions, sl_multiplier, pt_multiplier, horizon_days, cooldown_days, sizing_mode, risk_fraction, transaction_pct, account_mode, withdrawal_fraction, use_regime_routing, benchmark, use_confluence, min_confluence_score, ticker_preset, custom_tickers, progress=gr.Progress(), ): def _cb(msg, frac=None): progress(frac or 0, desc=msg) try: # ---- Load model ---- _cb("Resolving model artifacts from HuggingFace...", 0.01) bundle = _get_or_load_bundle(model_label, progress_cb=_cb) if bundle is None or bundle.main_model is None: return _bt_empty("❌ Could not load model. Check model label and connectivity.") # ---- Determine universe ---- tickers = _universe_for_preset(ticker_preset, custom_tickers, bundle) if not tickers: return _bt_empty("❌ No tickers found for selected universe.") # Also add benchmark ticker for download bm_sym = benchmark if benchmark != "None" else None extra = [bm_sym] if bm_sym else [] # ---- Download data ---- _cb("Downloading market data — batch 1/N...", 0.05) ticker_data = download_ticker_batch( tickers + extra, start=str(bt_start), end=str(bt_end), progress_cb=_cb, ) ticker_data = filter_ticker_data(ticker_data, progress_cb=_cb) # ---- Build config ---- config = BacktestConfig( start_date=str(bt_start), end_date=str(bt_end), initial_cash=float(initial_cash), conviction_threshold=float(conviction_threshold), use_regime_routing=bool(use_regime_routing), max_positions=int(max_positions), pt_multiplier=float(pt_multiplier), sl_multiplier=float(sl_multiplier), atr_period=14, horizon_days=int(horizon_days), cooldown_days=int(cooldown_days), sizing_mode="volatility_adjusted" if sizing_mode == "Volatility-adjusted" else "equal_weight", risk_fraction=float(risk_fraction) / 100.0, transaction_pct=float(transaction_pct) / 100.0, account_mode="compound" if account_mode == "Compound" else "realistic", withdrawal_fraction=float(withdrawal_fraction) / 100.0, use_confluence=bool(use_confluence), min_confluence_score=int(min_confluence_score), benchmark=benchmark, ) # ---- Run backtest ---- result = run_backtest(ticker_data, bundle, config, progress_cb=_cb) # ---- Build outputs ---- _cb("Rendering results...", 0.97) m = result.metrics metrics_html = _build_metrics_html(m) fig_nav = nav_chart(result.nav_df, result.benchmark_df, config.initial_cash) fig_monthly = monthly_returns_heatmap(result.nav_df, config.initial_cash) fig_exit = exit_reasons_chart(result.trades_df) fig_dist = trade_return_distribution(result.trades_df) trades_display = result.trades_df if not result.trades_df.empty else pd.DataFrame( columns=["Ticker","Entry Date","Exit Date","Exit Reason","Entry Price","Exit Price","Return %","Profit $","Days Held"] ) # CSV export string csv_str = result.trades_df.to_csv(index=False) if not result.trades_df.empty else "" json_str = json.dumps(result.metrics, indent=2, default=str) _cb("Done.", 1.0) return ( metrics_html, fig_nav, fig_monthly, fig_exit, fig_dist, trades_display, csv_str, json_str, f"✅ Backtest complete — {m.get('Total Trades', 0)} trades over {result.n_tickers_processed} tickers.", ) except Exception as e: logger.exception("Backtest error") return _bt_empty(f"❌ Error: {e}") def _bt_empty(msg): empty_fig = _empty_fig(msg) empty_df = pd.DataFrame() return "
%s
" % msg, \ empty_fig, empty_fig, empty_fig, empty_fig, empty_df, "", "{}", msg def _empty_fig(msg="No data"): import plotly.graph_objects as go fig = go.Figure() fig.update_layout( paper_bgcolor="#0d0f14", plot_bgcolor="#14171f", font=dict(color="#7a7f94"), annotations=[dict(text=msg, xref="paper", yref="paper", x=0.5, y=0.5, showarrow=False, font=dict(size=13, color="#7a7f94"))], height=300, ) return fig def _build_metrics_html(m: dict) -> str: if not m: return "

No metrics available.

" def _card(label, value, color="#e4e6ef", sub=""): return f"""
{label}
{value}
{f'
{sub}
' if sub else ''}
""" total_ret = m.get("Total Return %", 0) ann_ret = m.get("Annualized Return %", 0) sharpe = m.get("Sharpe Ratio", 0) max_dd = m.get("Max Drawdown %", 0) win_rate = m.get("Win Rate %", 0) n_trades = m.get("Total Trades", 0) pf = m.get("Profit Factor", 0) calmar = m.get("Calmar Ratio", 0) ret_color = "#00d4aa" if total_ret >= 0 else "#ff4d6a" dd_color = "#ff4d6a" if max_dd < -10 else "#f5a623" if max_dd < -5 else "#e4e6ef" sharpe_color = "#00d4aa" if sharpe > 1 else "#f5a623" if sharpe > 0.5 else "#ff4d6a" cards = "".join([ _card("Total Return", f"{total_ret:+.1f}%", ret_color), _card("Ann. Return", f"{ann_ret:+.1f}%", ret_color), _card("Sharpe", f"{sharpe:.2f}", sharpe_color), _card("Max Drawdown", f"{max_dd:.1f}%", dd_color), _card("Win Rate", f"{win_rate:.1f}%"), _card("Trades", f"{n_trades:,}"), _card("Profit Factor", f"{pf:.2f}"), _card("Calmar", f"{calmar:.2f}"), ]) return f"""
{cards}
""" def apply_preset(preset_name): cfg = PRESETS.get(preset_name, PRESETS["Balanced"]) sizing_label = "Volatility-adjusted" if cfg.sizing_mode == "volatility_adjusted" else "Equal-weight" account_label = "Compound" if cfg.account_mode == "compound" else "Realistic" return ( cfg.conviction_threshold, cfg.max_positions, cfg.sl_multiplier, cfg.pt_multiplier, cfg.horizon_days, cfg.cooldown_days, sizing_label, cfg.risk_fraction * 100, account_label, ) # --------------------------------------------------------------------------- # Evaluator runner # --------------------------------------------------------------------------- def run_evaluator( model_labels_selected, eval_start, eval_end, ticker_preset, custom_tickers, pt_multiplier, sl_multiplier, horizon, w_discrimination, w_feature_health, w_signal_stability, w_calibration, w_regime_robustness, w_asymmetry, progress=gr.Progress(), ): global _eval_results, _eval_labels def _cb(msg, frac=None): progress(frac or 0, desc=msg) if not model_labels_selected: return _eval_empty("❌ Select at least one model.") if isinstance(model_labels_selected, str): model_labels_selected = [model_labels_selected] weights = { "discrimination": w_discrimination / 100, "feature_health": w_feature_health / 100, "signal_stability": w_signal_stability / 100, "calibration": w_calibration / 100, "regime_robustness": w_regime_robustness / 100, "asymmetry": w_asymmetry / 100, } all_results = [] all_labels = [] for model_label in model_labels_selected: try: _cb(f"Loading model: {model_label}...", 0.02) bundle = _get_or_load_bundle(model_label, progress_cb=_cb) if bundle is None or bundle.main_model is None: logger.warning(f"Could not load {model_label}, skipping.") continue tickers = _universe_for_preset(ticker_preset, custom_tickers, bundle) if not tickers: continue _cb(f"Downloading data for {model_label}...", 0.05) ticker_data = download_ticker_batch( tickers, start=str(eval_start), end=str(eval_end), progress_cb=_cb ) ticker_data = filter_ticker_data(ticker_data, progress_cb=_cb) _cb(f"Evaluating {model_label}...", 0.38) result = run_evaluation( ticker_data=ticker_data, bundle=bundle, pt_multiplier=float(pt_multiplier), sl_multiplier=float(sl_multiplier), horizon=int(horizon), dimension_weights=weights, progress_cb=_cb, ) all_results.append(result) all_labels.append(model_label.split("·")[0].strip()) except Exception as e: logger.exception(f"Evaluation error for {model_label}") continue if not all_results: return _eval_empty("❌ No results produced. Check model and data.") _eval_results = all_results _eval_labels = all_labels primary = all_results[0] primary_label = all_labels[0] # Build grade HTML grade_html = _build_grade_html(primary, primary_label) # Build dim scores table dim_table = _build_dim_table(primary) # Charts fig_radar = radar_chart(primary.dimensions) fig_reliability = reliability_diagram(primary.reliability_bins) fig_regime = regime_heatmap(primary.regime_scores) fig_psi = feature_psi_chart(primary.feature_psi) fig_compare = multi_model_comparison(all_results, all_labels) if len(all_results) > 1 else _empty_fig("Add more models to enable comparison") # Flags all_flags = [] for dim in primary.dimensions: for flag in dim.flags: all_flags.append(f"⚠️ [{dim.name.replace('_',' ').title()}] {flag}") flags_html = _build_flags_html(all_flags) # PSI table psi_display = primary.feature_psi if not primary.feature_psi.empty else pd.DataFrame( columns=["Feature", "NaN Rate", "Inf Rate", "PSI", "Status"] ) status = f"✅ Evaluated {len(all_results)} model(s). Primary score: {primary.overall_score:.1f} ({primary.grade})" _cb("Done.", 1.0) return ( grade_html, dim_table, fig_radar, fig_reliability, fig_regime, fig_psi, fig_compare, psi_display, flags_html, status, ) def _build_grade_html(result, label: str) -> str: grade = result.grade score = result.overall_score color = GRADE_COLORS.get(grade, "#e4e6ef") dr = result.eval_date_range return f"""
{grade}
{score:.1f} / 100
{label}
{result.n_samples:,} samples · {result.n_positives} positives ({result.n_positives/max(result.n_samples,1):.1%} rate) · {dr[0]} → {dr[1]}
""" def _build_dim_table(result) -> str: rows = "" for dim in result.dimensions: bar_color = "#00d4aa" if dim.score >= 70 else "#f5a623" if dim.score >= 50 else "#ff4d6a" bar_w = int(dim.score) label = dim.name.replace("_", " ").title() rows += f""" {label}
{dim.score:.1f} {int(dim.weight*100)}% weight """ return f""" {rows}
Dimension Score Bar Score Weight
""" def _build_flags_html(flags: list) -> str: if not flags: return "
✅ No critical issues detected.
" items = "".join( f"
  • {f}
    • " for f in flags ) return f"" def _eval_empty(msg): empty_fig = _empty_fig(msg) empty_df = pd.DataFrame() err_html = f"
    {msg}
    " return err_html, err_html, empty_fig, empty_fig, empty_fig, empty_fig, empty_fig, empty_df, err_html, msg # --------------------------------------------------------------------------- # CSS # --------------------------------------------------------------------------- CUSTOM_CSS = """ @import url('https://fonts.googleapis.com/css2?family=DM+Sans:wght@400;500;600;700&family=JetBrains+Mono:wght@400;600;700&display=swap'); :root { --bg: #0d0f14; --surface: #14171f; --surface2: #1c202c; --border: #2a2f3d; --text: #e4e6ef; --muted: #7a7f94; --teal: #00d4aa; --red: #ff4d6a; --amber: #f5a623; --purple: #a78bfa; } body, .gradio-container { background: var(--bg) !important; font-family: 'DM Sans', sans-serif !important; color: var(--text) !important; } .gradio-container { max-width: 1400px !important; } /* Header */ .sniper-header { padding: 28px 0 20px; border-bottom: 1px solid var(--border); margin-bottom: 24px; } .sniper-header h1 { font-size: 28px; font-weight: 700; color: var(--text); margin: 0; display: flex; align-items: center; gap: 10px; } .sniper-header p { color: var(--muted); font-size: 13px; margin: 6px 0 0; } /* Tab styling */ .tab-nav button { font-family: 'DM Sans', sans-serif !important; font-size: 14px !important; font-weight: 600 !important; color: var(--muted) !important; background: transparent !important; border-bottom: 2px solid transparent !important; padding: 10px 20px !important; transition: all 0.2s ease !important; } .tab-nav button.selected { color: var(--teal) !important; border-bottom: 2px solid var(--teal) !important; } /* Panel backgrounds */ .panel-box { background: var(--surface); border: 1px solid var(--border); border-radius: 12px; padding: 20px; } /* Sliders */ input[type=range] { accent-color: var(--teal) !important; } /* Labels */ label span { color: var(--text) !important; font-size: 13px !important; font-weight: 500 !important; } /* Inputs */ input, select, textarea { background: var(--surface2) !important; border: 1px solid var(--border) !important; color: var(--text) !important; border-radius: 8px !important; } input:focus, select:focus, textarea:focus { border-color: var(--teal) !important; outline: none !important; box-shadow: 0 0 0 2px rgba(0, 212, 170, 0.15) !important; } /* Buttons */ .btn-primary { background: var(--teal) !important; color: #0d0f14 !important; font-weight: 700 !important; font-family: 'DM Sans', sans-serif !important; border: none !important; border-radius: 8px !important; padding: 12px 28px !important; font-size: 14px !important; cursor: pointer !important; transition: opacity 0.2s ease !important; width: 100% !important; } .btn-primary:hover { opacity: 0.88 !important; } .btn-secondary { background: var(--surface2) !important; color: var(--text) !important; font-weight: 600 !important; font-family: 'DM Sans', sans-serif !important; border: 1px solid var(--border) !important; border-radius: 8px !important; padding: 10px 20px !important; font-size: 13px !important; cursor: pointer !important; width: 100% !important; } /* Progress bar */ .progress-bar { background: var(--teal) !important; } /* Status box */ .status-box { background: var(--surface); border: 1px solid var(--border); border-radius: 8px; padding: 12px 16px; font-family: 'JetBrains Mono', monospace; font-size: 12px; color: var(--muted); min-height: 48px; } /* Dataframe */ .dataframe { background: var(--surface) !important; border: 1px solid var(--border) !important; border-radius: 10px !important; overflow: hidden !important; } .dataframe th { background: var(--surface2) !important; color: var(--muted) !important; font-size: 11px !important; text-transform: uppercase !important; letter-spacing: 0.06em !important; border-bottom: 1px solid var(--border) !important; padding: 10px 12px !important; } .dataframe td { color: var(--text) !important; font-family: 'JetBrains Mono', monospace !important; font-size: 12px !important; padding: 9px 12px !important; border-bottom: 1px solid var(--border) !important; } /* Section labels */ .section-label { font-size: 11px; font-weight: 600; text-transform: uppercase; letter-spacing: 0.08em; color: var(--muted); margin-bottom: 8px; } /* Preset radio */ .preset-radio label { display: inline-flex !important; align-items: center !important; gap: 6px !important; padding: 7px 14px !important; border: 1px solid var(--border) !important; border-radius: 6px !important; cursor: pointer !important; font-size: 13px !important; transition: all 0.15s ease !important; } .preset-radio label:has(input:checked) { border-color: var(--teal) !important; color: var(--teal) !important; background: rgba(0,212,170,0.07) !important; } /* Accordion */ .accordion { border: 1px solid var(--border) !important; border-radius: 10px !important; background: var(--surface) !important; } /* Dividers */ hr { border-color: var(--border) !important; } /* Refresh button */ .refresh-row { margin-bottom: 20px; } """ # --------------------------------------------------------------------------- # Build Gradio UI # --------------------------------------------------------------------------- def build_ui(): with gr.Blocks(css=CUSTOM_CSS, title="🎯 Sniper Model Evaluator") as demo: # ---- Header ---- gr.HTML("""

    🎯 Sniper Model Evaluator

    Asymmetric precision stock signal models — backtesting & diagnostic suite

    """) # ---- Registry row ---- with gr.Row(elem_classes="refresh-row"): with gr.Column(scale=5): registry_status = gr.Textbox( label="Registry Status", value="Click 'Refresh Models' to discover available model runs.", interactive=False, elem_classes="status-box", lines=2, ) with gr.Column(scale=1): refresh_btn = gr.Button("🔄 Refresh Models", elem_classes="btn-secondary") # ---- Tabs ---- with gr.Tabs(): # ============================================================== # TAB 1 — BACKTESTER # ============================================================== with gr.Tab("📈 Backtester"): with gr.Row(): # ---- LEFT: Parameter panel ---- with gr.Column(scale=1, min_width=310): gr.HTML('
    Model
    ') bt_model = gr.Dropdown( choices=[], value=None, label="Model / Run", info="Auto-discovered from Arkm20/sniper-model-*", ) gr.HTML('
    Preset
    ') preset_radio = gr.Radio( choices=["Conservative", "Balanced", "Aggressive", "Paper test"], value="Balanced", label="", elem_classes="preset-radio", ) gr.HTML('
    Date Range
    ') with gr.Row(): bt_start = gr.Textbox(value="2021-01-01", label="Start date", max_lines=1) bt_end = gr.Textbox(value="2024-12-31", label="End date", max_lines=1) gr.HTML('
    Capital & Signal
    ') initial_cash = gr.Number(value=10000, label="Initial capital ($)", minimum=1000, maximum=1_000_000, step=1000) conviction_thr = gr.Slider(0.20, 0.90, value=0.50, step=0.01, label="Conviction threshold", info="Minimum calibrated probability to enter a trade") gr.HTML('
    Position Controls
    ') max_positions = gr.Slider(1, 20, value=5, step=1, label="Max concurrent positions") sl_mult = gr.Slider(0.25, 3.0, value=0.5, step=0.05, label="Stop loss (× ATR)", info="Mirrors trainer's SL_MULTIPLIER") pt_mult = gr.Slider(1.0, 6.0, value=3.0, step=0.1, label="Profit target (× ATR)", info="Mirrors trainer's PT_MULTIPLIER") horizon_days = gr.Slider(5, 60, value=15, step=1, label="Max hold horizon (days)") cooldown_days = gr.Slider(0, 10, value=2, step=1, label="Cooldown after exit (days)") gr.HTML('
    Sizing
    ') sizing_mode = gr.Radio( ["Volatility-adjusted", "Equal-weight"], value="Volatility-adjusted", label="Position sizing mode", ) risk_fraction = gr.Slider(0.5, 5.0, value=2.0, step=0.1, label="Risk % of NAV per trade", info="Used in volatility-adjusted mode only") with gr.Accordion("Advanced", open=False): transaction_pct = gr.Slider(0.0, 0.5, value=0.1, step=0.01, label="Transaction cost (%)", info="Applied on entry and exit") account_mode = gr.Radio( ["Compound", "Realistic"], value="Compound", label="Account mode", info="Realistic withdraws a fraction of profits to personal cash", ) withdrawal_fraction = gr.Slider(5, 50, value=20, step=5, label="Profit withdrawal % (Realistic mode)") use_regime = gr.Checkbox(value=True, label="Auto-route to regime models") benchmark = gr.Dropdown(["SPY", "QQQ", "None"], value="SPY", label="Benchmark overlay") use_confluence = gr.Checkbox(value=False, label="Confluence filter") min_confluence = gr.Slider(1, 10, value=3, step=1, label="Min confluence score") gr.HTML('
    Ticker Universe
    ') ticker_preset = gr.Dropdown( choices=list(TICKER_UNIVERSE_PRESETS.keys()), value="All (~800 tickers)", label="Universe", ) custom_tickers = gr.Textbox( label="Custom tickers (space or comma separated)", placeholder="AAPL MSFT NVDA TSLA ...", visible=False, lines=3, ) bt_run_btn = gr.Button("▶ Run Backtest", elem_classes="btn-primary", variant="primary") bt_status = gr.Textbox(label="Status", value="", interactive=False, elem_classes="status-box", lines=1) # ---- RIGHT: Results panel ---- with gr.Column(scale=3): bt_metrics_html = gr.HTML( "
    " "Configure parameters and run a backtest to see results.
    " ) bt_nav_plot = gr.Plot(label="") with gr.Row(): bt_monthly_plot = gr.Plot(label="") bt_exit_plot = gr.Plot(label="") bt_dist_plot = gr.Plot(label="") with gr.Accordion("Trade Log", open=False): bt_trades_df = gr.Dataframe( headers=["Ticker","Entry Date","Exit Date","Exit Reason", "Entry Price","Exit Price","Return %","Profit $","Days Held"], label="", wrap=False, ) with gr.Accordion("Export", open=False): with gr.Row(): bt_csv_out = gr.Textbox(label="Trades CSV", lines=6, interactive=False) bt_json_out = gr.Textbox(label="Metrics JSON", lines=6, interactive=False) # ============================================================== # TAB 2 — EVALUATOR # ============================================================== with gr.Tab("🔬 Model Evaluator"): with gr.Row(): # ---- LEFT: Config ---- with gr.Column(scale=1, min_width=310): gr.HTML('
    Models to Evaluate
    ') eval_models = gr.Dropdown( choices=[], value=None, multiselect=True, label="Model runs (select one or more for comparison)", ) gr.HTML('
    Evaluation Range
    ') with gr.Row(): eval_start = gr.Textbox(value="2019-01-01", label="Start date", max_lines=1) eval_end = gr.Textbox(value="2020-12-31", label="End date", max_lines=1) gr.HTML('
    Label Parameters
    ') gr.HTML('

    Match these to the model\'s training config for accurate scoring.

    ') eval_pt = gr.Slider(1.0, 6.0, value=3.0, step=0.1, label="Profit target (× ATR)", info="Should match PT_MULTIPLIER in training") eval_sl = gr.Slider(0.1, 3.0, value=0.5, step=0.05, label="Stop loss (× ATR)", info="Should match SL_MULTIPLIER in training") eval_horizon = gr.Slider(5, 60, value=15, step=1, label="Horizon (days)", info="Should match HORIZON in training") gr.HTML('
    Ticker Universe
    ') eval_ticker_preset = gr.Dropdown( choices=list(TICKER_UNIVERSE_PRESETS.keys()), value="S&P 500 subset (200)", label="Universe", info="Smaller universe = faster evaluation", ) eval_custom_tickers = gr.Textbox( label="Custom tickers", placeholder="AAPL MSFT NVDA ...", visible=False, lines=3, ) with gr.Accordion("Scoring Weights", open=False): gr.HTML('

    Adjust how each dimension contributes to the overall score. Values are relative.

    ') w_disc = gr.Slider(0, 100, value=20, step=5, label="Discrimination") w_feat = gr.Slider(0, 100, value=20, step=5, label="Feature Health") w_stab = gr.Slider(0, 100, value=15, step=5, label="Signal Stability") w_cal = gr.Slider(0, 100, value=15, step=5, label="Calibration") w_reg = gr.Slider(0, 100, value=15, step=5, label="Regime Robustness") w_asym = gr.Slider(0, 100, value=15, step=5, label="Asymmetry Capture") eval_run_btn = gr.Button("▶ Run Evaluation", elem_classes="btn-primary", variant="primary") eval_status = gr.Textbox(label="Status", value="", interactive=False, elem_classes="status-box", lines=1) # ---- RIGHT: Results ---- with gr.Column(scale=3): eval_grade_html = gr.HTML( "
    " "Select models and run evaluation to see the scorecard.
    " ) eval_dim_table = gr.HTML("") with gr.Row(): eval_radar = gr.Plot(label="") eval_reliability = gr.Plot(label="") with gr.Row(): eval_regime = gr.Plot(label="") eval_psi_chart = gr.Plot(label="") eval_compare = gr.Plot(label="") with gr.Accordion("Feature PSI Table", open=False): eval_psi_table = gr.Dataframe( headers=["Feature", "NaN Rate", "Inf Rate", "PSI", "Status"], label="", wrap=False, ) with gr.Accordion("Warnings & Flags", open=False): eval_flags_html = gr.HTML("") # --------------------------------------------------------------------------- # Event wiring # --------------------------------------------------------------------------- # Refresh registry → update both dropdowns refresh_btn.click( fn=_refresh_registry, outputs=[bt_model, eval_models, registry_status], ) # Preset → fill sliders preset_radio.change( fn=apply_preset, inputs=[preset_radio], outputs=[conviction_thr, max_positions, sl_mult, pt_mult, horizon_days, cooldown_days, sizing_mode, risk_fraction, account_mode], ) # Show custom ticker box when "Custom" selected def _toggle_custom_bt(val): return gr.Textbox(visible=(val == "Custom (paste tickers)")) def _toggle_custom_eval(val): return gr.Textbox(visible=(val == "Custom (paste tickers)")) ticker_preset.change(fn=_toggle_custom_bt, inputs=[ticker_preset], outputs=[custom_tickers]) eval_ticker_preset.change(fn=_toggle_custom_eval, inputs=[eval_ticker_preset], outputs=[eval_custom_tickers]) # Run backtest bt_run_btn.click( fn=run_backtester, inputs=[ bt_model, bt_start, bt_end, initial_cash, conviction_thr, max_positions, sl_mult, pt_mult, horizon_days, cooldown_days, sizing_mode, risk_fraction, transaction_pct, account_mode, withdrawal_fraction, use_regime, benchmark, use_confluence, min_confluence, ticker_preset, custom_tickers, ], outputs=[ bt_metrics_html, bt_nav_plot, bt_monthly_plot, bt_exit_plot, bt_dist_plot, bt_trades_df, bt_csv_out, bt_json_out, bt_status, ], ) # Run evaluator eval_run_btn.click( fn=run_evaluator, inputs=[ eval_models, eval_start, eval_end, eval_ticker_preset, eval_custom_tickers, eval_pt, eval_sl, eval_horizon, w_disc, w_feat, w_stab, w_cal, w_reg, w_asym, ], outputs=[ eval_grade_html, eval_dim_table, eval_radar, eval_reliability, eval_regime, eval_psi_chart, eval_compare, eval_psi_table, eval_flags_html, eval_status, ], ) # Auto-refresh on load demo.load(fn=_refresh_registry, outputs=[bt_model, eval_models, registry_status]) return demo # --------------------------------------------------------------------------- # Entry point # --------------------------------------------------------------------------- if __name__ == "__main__": app = build_ui() app.launch( server_name="0.0.0.0", server_port=7860, show_error=True, )