Upload Quasar_axrvi_ranker.py

Quasar_axrvi_ranker.py

@@ -368,8 +368,6 @@ DERIV_WS_URL   = "wss://ws.binaryws.com/websockets/v3?app_id=1089"
 # Deriv API symbol → AXRVI internal symbol
 SYMBOL_MAP = {
     "R_25":      "V25",
-    "1HZ30V":    "V30_1s",
-    "R_50":      "V50",
     "1HZ50V":    "V50_1s",
     "R_75":      "V75",
     "1HZ75V":    "V75_1s",
@@ -398,8 +396,6 @@ del _reverse_check, _deriv_sym, _axrvi_id
 # Per-asset metadata: base volatility and max position fraction
 ASSET_REGISTRY: Dict[str, dict] = {
     "V25":      {"symbol": "R_25",      "base_vol": 25.0,  "max_pos": 0.006},
-    "V30_1s":   {"symbol": "1HZ30V",    "base_vol": 30.0,  "max_pos": 0.004},
-    "V50":      {"symbol": "R_50",      "base_vol": 50.0,  "max_pos": 0.005},
     "V50_1s":   {"symbol": "1HZ50V",    "base_vol": 50.0,  "max_pos": 0.004},
     "V75":      {"symbol": "R_75",      "base_vol": 75.0,  "max_pos": 0.005},
     "V75_1s":   {"symbol": "1HZ75V",    "base_vol": 75.0,  "max_pos": 0.003},
@@ -423,8 +419,6 @@ ASSET_REGISTRY: Dict[str, dict] = {
 #
 # Accepted ranges per Deriv broker:
 #   V25       — [160, 400, 800, 1200, 1600] → use 160
-#   V30_1s    — [140, 400, 700, 1000, 1400] → use 140
-#   V50       — [80, 200, 400, 600, 800] → use 80   ← FIXED: broker confirmed, NOT [50,100,200,300,500]
 #   V50_1s    — [80, 200, 400, 600, 800] → use 80
 #   V75       — [50, 100, 200, 300, 500] → use 50
 #   V75_1s    — [50, 100, 200, 300, 500] → use 50
@@ -436,8 +430,6 @@ ASSET_REGISTRY: Dict[str, dict] = {
 #   STEP200   — [400, 1000, 2000, 3000, 4000] → use 400
 ASSET_MULTIPLIER: Dict[str, int] = {
     "V25":      160,   # FIXED: was 50 → rejected, now 160 ✓
-    "V30_1s":   140,    # smallest accepted from [140, 400, 700, 1000, 1400] ✓
-    "V50":      80,    # FIXED: was 50 → rejected (broker accepts 80,200,400,600,800), now 80 ✓
     "V50_1s":   80,    # FIXED: was 30 → rejected, now 80 ✓
     "V75":      50,    # FIXED: was 30 → rejected, now 50 ✓
     "V75_1s":   50,    # FIXED: was 20 → rejected, now 50 ✓
@@ -452,8 +444,6 @@ ASSET_MULTIPLIER: Dict[str, int] = {
 # ── Broker's acceptable multiplier ranges (for validation & future fallback) ──
 ASSET_ACCEPTABLE_MULTIPLIERS: Dict[str, List[int]] = {
     "V25":      [160, 400, 800, 1200, 1600],
-    "V30_1s":   [140, 400, 700, 1000, 1400],  # confirmed: 140 is lowest accepted
-    "V50":      [80, 200, 400, 600, 800],  # FIXED: broker confirmed NOT [50,100,200,300,500]
     "V50_1s":   [80, 200, 400, 600, 800],
     "V75":      [50, 100, 200, 300, 500],
     "V75_1s":   [50, 100, 200, 300, 500],
@@ -469,8 +459,6 @@ ASSET_ACCEPTABLE_MULTIPLIERS: Dict[str, List[int]] = {
 # e.g. 0.50 = close when $0.50 of the $1 stake is lost
 ASSET_STOP_LOSS_FRAC: Dict[str, float] = {
     "V25":      0.60,
-    "V30_1s":   0.55,
-    "V50":      0.55,
     "V50_1s":   0.55,
     "V75":      0.50,
     "V75_1s":   0.45,
@@ -485,8 +473,6 @@ ASSET_STOP_LOSS_FRAC: Dict[str, float] = {
 # Take-profit as fraction of stake (exit early when profit target hit)
 ASSET_TAKE_PROFIT_FRAC: Dict[str, float] = {
     "V25":      1.00,
-    "V30_1s":   0.90,
-    "V50":      0.90,
     "V50_1s":   0.90,
     "V75":      0.80,
     "V75_1s":   0.75,
@@ -4620,8 +4606,8 @@ def create_axrvi_v8(num_assets: int = 5,
 
     Example::
 
-        model = create_axrvi_v8(num_assets=9, config=cfg, device="cpu")
-        out   = model(sequences)            # (1, 9, 20, 26)
+        model = create_axrvi_v8(num_assets=10, config=cfg, device="cpu")
+        out   = model(sequences)            # (1, 10, 20, 26)
         loss  = v8_total_loss(out, rank_t, returns, model.distributional.quantile_levels)
     """
     if config is None:
@@ -6813,9 +6799,9 @@ class QuasarAXRVIBridge:
         reward_strategy: str  = "simple",
         hub_ws_url:      str  = os.environ.get("QUASAR_HUB_URL", "ws://localhost:7860/ws/subscribe"),
         enable_logging:  bool = True,
-        checkpoint_dir:  str  = "./Ranker8",   # FIX: fresh folder
-        resume:          bool = False,         # FIX: fresh start
-        hf_repo_id:      Optional[str] = "KarlQuant/k1rl-checkpoints",  # HF Dataset repo
+        checkpoint_dir:  str  = "./Ranker10",  # new folder for 10-asset build
+        resume:          bool = False,          # FIX: fresh start
+        hf_repo_id:      Optional[str] = "KarlQuant/quasar-axrvi-v10",  # new HF repo (10 assets)
     ):
         self.config          = config or AssetRankerConfig()
         self.trade_config    = trade_config or TradeConfig()
@@ -6825,7 +6811,7 @@ class QuasarAXRVIBridge:
         # ── Checkpoint manager (local + optional HF sync) ─────────────────────
         self.checkpoint_mgr  = RankerCheckpointManager(
             checkpoint_dir=checkpoint_dir,
-            hf_repo_id=hf_repo_id,          # "KarlQuant/k1rl-checkpoints"
+            hf_repo_id=hf_repo_id,
         )
         self.resume          = resume
 
@@ -8737,6 +8723,22 @@ class HFDatasetCheckpointManager:
     def _hf_ckpt_path(self, step: int) -> str:
         return f"step_{step:07d}.pt"
 
+    def _ensure_repo_exists(self) -> None:
+        """Create the HF Dataset repo if it does not already exist. No-op if it exists."""
+        if not self._ensure_hf():
+            return
+        try:
+            self._hfapi.create_repo(
+                repo_id   = self.repo_id,
+                repo_type = "dataset",
+                exist_ok  = True,   # safe to call even if repo already exists
+                private   = True,
+            )
+            if self.verbose:
+                logger.info(f"☁️  HF repo ready (created or already exists): {self.repo_id}")
+        except Exception as exc:
+            logger.warning(f"☁️  Could not ensure HF repo exists (non-fatal): {exc}")
+
     # ── Upload ─────────────────────────────────────────────────────────────────
 
     def upload(
@@ -8758,6 +8760,9 @@ class HFDatasetCheckpointManager:
         if not self._ensure_hf():
             return False
 
+        # Auto-create the repo if this is a new HF folder (no-op if it already exists)
+        self._ensure_repo_exists()
+
         local_path = Path(local_path)
         if not local_path.exists():
             logger.warning(f"☁️  HF upload skipped — file not found: {local_path}")
@@ -9078,7 +9083,7 @@ class RankerCheckpointManager:
 
     def __init__(
         self,
-        checkpoint_dir: str = "./Ranker8",
+        checkpoint_dir: str = "./Ranker10",
         hf_repo_id: Optional[str] = None,
     ):
         self.checkpoint_dir  = checkpoint_dir
@@ -9245,10 +9250,12 @@ class RankerCheckpointManager:
     def _build_checkpoint(self, bridge, step: int, reason: str) -> dict:
         """Collect all live bridge state into a flat checkpoint dict."""
         ckpt: dict = {
-            "version": "2.0",
-            "step":    step,
-            "reason":  reason,
-            "timestamp": datetime.now().isoformat(),
+            "version":    "2.1",
+            "step":       step,
+            "reason":     reason,
+            "timestamp":  datetime.now().isoformat(),
+            "num_assets": bridge.axrvi_net.num_assets if bridge.axrvi_net is not None else 0,
+            "asset_symbols": list(bridge.config.asset_symbols),
         }
 
         # ── Model ─────────────────────────────────────────────────────────────
@@ -9306,22 +9313,48 @@ class RankerCheckpointManager:
         # ── Validate critical keys ─────────────────────────────────────────────
         assert "step" in ckpt, "Checkpoint missing 'step' key"
 
+        # ── num_assets compatibility guard ────────────────────────────────────
+        # If the checkpoint was saved with a different asset count (e.g. 12 before
+        # V50/V30_1s were removed), the QCSAM/FABLE weight tensors will have the
+        # wrong shape. We detect this early and skip model + optimizer weights so
+        # the rest of the state (replay, bandit, counters) can still be restored.
+        ckpt_num_assets    = ckpt.get("num_assets", -1)
+        current_num_assets = bridge.axrvi_net.num_assets if bridge.axrvi_net is not None else -1
+        _model_compatible  = True
+
+        if ckpt_num_assets != -1 and ckpt_num_assets != current_num_assets:
+            ckpt_assets = ckpt.get("asset_symbols", "unknown")
+            logger.warning(
+                f"⚠️  [Restore] Asset count mismatch: checkpoint has {ckpt_num_assets} assets "
+                f"({ckpt_assets}), but current model has {current_num_assets} assets "
+                f"({list(bridge.config.asset_symbols)}). "
+                f"Skipping axrvi_net + optimizer weights — model starts fresh. "
+                f"All other state (replay, bandit, counters) will be restored."
+            )
+            _model_compatible = False
+
         # ── Model ─────────────────────────────────────────────────────────────
         if "axrvi_net" in ckpt and bridge.axrvi_net is not None:
-            incompatible = bridge.axrvi_net.load_state_dict(ckpt["axrvi_net"], strict=False)
-            if incompatible.missing_keys:
-                logger.warning(f"[Restore] axrvi_net missing keys: {incompatible.missing_keys}")
-            if incompatible.unexpected_keys:
-                logger.warning(f"[Restore] axrvi_net unexpected keys: {incompatible.unexpected_keys}")
-            logger.info("   ✅ axrvi_net restored")
+            if _model_compatible:
+                incompatible = bridge.axrvi_net.load_state_dict(ckpt["axrvi_net"], strict=False)
+                if incompatible.missing_keys:
+                    logger.warning(f"[Restore] axrvi_net missing keys: {incompatible.missing_keys}")
+                if incompatible.unexpected_keys:
+                    logger.warning(f"[Restore] axrvi_net unexpected keys: {incompatible.unexpected_keys}")
+                logger.info("   ✅ axrvi_net restored")
+            else:
+                logger.info("   ⏭️  axrvi_net skipped (asset count mismatch — fresh weights kept)")
 
         # ── Trainer ───────────────────────────────────────────────────────────
         if bridge.trainer is not None:
             tr = bridge.trainer
-            if "optimizer" in ckpt:
-                tr.optimizer.load_state_dict(ckpt["optimizer"])
-            if "scheduler" in ckpt:
-                tr.scheduler.load_state_dict(ckpt["scheduler"])
+            if _model_compatible:
+                if "optimizer" in ckpt:
+                    tr.optimizer.load_state_dict(ckpt["optimizer"])
+                if "scheduler" in ckpt:
+                    tr.scheduler.load_state_dict(ckpt["scheduler"])
+            else:
+                logger.info("   ⏭️  optimizer/scheduler skipped (asset count mismatch — fresh state kept)")
             tr.train_step   = ckpt.get("train_step",   ckpt.get("step", tr.train_step))
             tr.lambda_ce    = ckpt.get("lambda_ce",    tr.lambda_ce)
             tr.lambda_ql    = ckpt.get("lambda_ql",    tr.lambda_ql)
@@ -10387,9 +10420,9 @@ async def run_live_trading_system(
     hub_ws_url:      str  = "ws://localhost:7860/ws/subscribe",
     enable_logging:  bool = True,
     shreve_config:   Optional[ShreveConfig] = None,
-    checkpoint_dir:  str  = "./Ranker8",
+    checkpoint_dir:  str  = "./Ranker10",
     resume:          bool = False,   # FIX: fresh start
-    hf_repo_id:      Optional[str] = "KarlQuant/k1rl-checkpoints",  # HF Dataset repo
+    hf_repo_id:      Optional[str] = "KarlQuant/quasar-axrvi-v10",  # new HF repo (10 assets)
 ) -> None:
     config = AssetRankerConfig(
         asset_symbols    = asset_symbols or list(ASSET_REGISTRY.keys()),
@@ -10701,12 +10734,12 @@ def _parse_args():
                         help="[S6/S8] Trade horizon τ in seconds (default 60)")
     parser.add_argument("--martingale-epsilon", type=float, default=0.05,
                         help="[S7] Gate E martingale deviation threshold (default 0.05)")
-    parser.add_argument("--checkpoint-dir", default="./Ranker8",
-                        help="Directory for full-state checkpoints (default ./Ranker7)")
+    parser.add_argument("--checkpoint-dir", default="./Ranker10",
+                        help="Directory for full-state checkpoints (default ./Ranker10)")
     parser.add_argument("--no-resume", dest="no_resume", action="store_true", default=True,
                         help="Default True — always fresh start.")
     parser.add_argument("--resume", dest="no_resume", action="store_false",
-                        help="Restore from latest Ranker7 checkpoint")
+                        help="Restore from latest Ranker10 checkpoint")
     parser.add_argument("--hf-repo", default=None,
                         metavar="OWNER/REPO",
                         help="Hugging Face Dataset repo for checkpoint sync "
@@ -10763,8 +10796,8 @@ if __name__ == "__main__":
         hub_ws_url      = args.hub,
         enable_logging  = not args.no_logs,
         checkpoint_dir  = args.checkpoint_dir,
-        resume          = not args.no_resume,   # default False — always start fresh on Ranker8
-        hf_repo_id      = args.hf_repo or "KarlQuant/k1rl-checkpoints",
+        resume          = not args.no_resume,   # default False — always start fresh on Ranker10
+        hf_repo_id      = args.hf_repo or "KarlQuant/quasar-axrvi-v10",
     )
 
     try:
@@ -10786,8 +10819,8 @@ if __name__ == "__main__":
                 hub_ws_url      = args.hub,
                 enable_logging  = not args.no_logs,
                 checkpoint_dir  = args.checkpoint_dir,   # FIX 1: was silently ignored
-                resume          = not args.no_resume,      # FIX 2: default True (always resume)
-                hf_repo_id      = args.hf_repo or "KarlQuant/k1rl-checkpoints",
+                resume          = not args.no_resume,      # FIX 2: default False (always fresh start)
+                hf_repo_id      = args.hf_repo or "KarlQuant/quasar-axrvi-v10",
             ))
     except KeyboardInterrupt:
         print("\n👋 Shutting down…")