update

example/export_samples.py

@@ -12,7 +12,7 @@ from typing import Dict, List
 import torch
 import torch.nn.functional as F
 
-from data_utils import load_split, inverse_quantile_transform, quantile_calibrate_to_real
+from data_utils import load_split, normalize_cont, inverse_quantile_transform, quantile_calibrate_to_real
 from hybrid_diffusion import HybridDiffusionModel, TemporalGRUGenerator, TemporalTransformerGenerator, cosine_beta_schedule
 from platform_utils import resolve_device, safe_path, ensure_dir, resolve_path
 
@@ -157,10 +157,15 @@ def main():
     cont_post_calibrate = bool(cfg.get("cont_post_calibrate", False))
     type1_cols = cfg.get("type1_features", []) or []
     type5_cols = cfg.get("type5_features", []) or []
+    type4_cols = cfg.get("type4_features", []) or []
     type1_cols = [c for c in type1_cols if c in cont_cols]
     type5_cols = [c for c in type5_cols if c in cont_cols]
+    type4_cols = [c for c in type4_cols if c in cont_cols]
     model_cont_cols = [c for c in cont_cols if c not in type1_cols and c not in type5_cols]
     use_temporal_stage1 = bool(cfg.get("use_temporal_stage1", False))
+    temporal_use_type1_cond = bool(cfg.get("temporal_use_type1_cond", False))
+    temporal_focus_type4 = bool(cfg.get("temporal_focus_type4", False))
+    temporal_exclude_type4 = bool(cfg.get("temporal_exclude_type4", False))
     temporal_backbone = str(cfg.get("temporal_backbone", "gru"))
     temporal_hidden_dim = int(cfg.get("temporal_hidden_dim", 256))
     temporal_num_layers = int(cfg.get("temporal_num_layers", 1))
@@ -207,6 +212,22 @@ def main():
 
     temporal_model = None
     if use_temporal_stage1:
+        temporal_path = Path(args.model_path).with_name("temporal.pt")
+        if not temporal_path.exists():
+            raise SystemExit(f"missing temporal model file: {temporal_path}")
+        temporal_state = load_torch_state(str(temporal_path), device)
+        temporal_cond_dim = len(type1_cols) if (temporal_use_type1_cond and type1_cols) else 0
+        if isinstance(temporal_state, dict):
+            if "in_proj.weight" in temporal_state:
+                try:
+                    temporal_cond_dim = max(0, int(temporal_state["in_proj.weight"].shape[1]) - len(model_cont_cols))
+                except Exception:
+                    pass
+            elif "gru.weight_ih_l0" in temporal_state:
+                try:
+                    temporal_cond_dim = max(0, int(temporal_state["gru.weight_ih_l0"].shape[1]) - len(model_cont_cols))
+                except Exception:
+                    pass
         if temporal_backbone == "transformer":
             temporal_model = TemporalTransformerGenerator(
                 input_dim=len(model_cont_cols),
@@ -217,6 +238,7 @@ def main():
                 dropout=temporal_transformer_dropout,
                 pos_dim=temporal_pos_dim,
                 use_pos_embed=temporal_use_pos_embed,
+                cond_dim=temporal_cond_dim,
             ).to(device)
         else:
             temporal_model = TemporalGRUGenerator(
@@ -224,11 +246,9 @@ def main():
                 hidden_dim=temporal_hidden_dim,
                 num_layers=temporal_num_layers,
                 dropout=temporal_dropout,
+                cond_dim=temporal_cond_dim,
             ).to(device)
-        temporal_path = Path(args.model_path).with_name("temporal.pt")
-        if not temporal_path.exists():
-            raise SystemExit(f"missing temporal model file: {temporal_path}")
-        temporal_model.load_state_dict(load_torch_state(str(temporal_path), device))
+        temporal_model.load_state_dict(temporal_state)
         temporal_model.eval()
 
     betas = cosine_beta_schedule(args.timesteps).to(device)
@@ -279,13 +299,32 @@ def main():
                     for t, row in enumerate(seq):
                         for i, c in enumerate(type1_cols):
                             cond_cont[:, t, i] = float(row[c])
-                    mean_vec = torch.tensor([mean[c] for c in type1_cols], dtype=cond_cont.dtype, device=device)
-                    std_vec = torch.tensor([std[c] for c in type1_cols], dtype=cond_cont.dtype, device=device)
-                    cond_cont = (cond_cont - mean_vec) / std_vec
+                    cond_cont = normalize_cont(
+                        cond_cont,
+                        type1_cols,
+                        mean,
+                        std,
+                        transforms=transforms,
+                        quantile_probs=quantile_probs,
+                        quantile_values=quantile_values,
+                        use_quantile=use_quantile,
+                    )
 
     trend = None
     if temporal_model is not None:
-        trend = temporal_model.generate(args.batch_size, args.seq_len, device)
+        trend = temporal_model.generate(args.batch_size, args.seq_len, device, cond_cont=cond_cont)
+        if temporal_focus_type4 and type4_cols:
+            type4_model_idx = [model_cont_cols.index(c) for c in type4_cols if c in model_cont_cols]
+            if type4_model_idx:
+                trend_mask = torch.zeros(1, 1, len(model_cont_cols), device=device, dtype=trend.dtype)
+                trend_mask[:, :, type4_model_idx] = 1.0
+                trend = trend * trend_mask
+        elif temporal_exclude_type4 and type4_cols:
+            type4_model_idx = [model_cont_cols.index(c) for c in type4_cols if c in model_cont_cols]
+            if type4_model_idx:
+                trend_mask = torch.ones(1, 1, len(model_cont_cols), device=device, dtype=trend.dtype)
+                trend_mask[:, :, type4_model_idx] = 0.0
+                trend = trend * trend_mask
 
     for t in reversed(range(args.timesteps)):
         t_batch = torch.full((args.batch_size,), t, device=device, dtype=torch.long)

example/hybrid_diffusion.py

@@ -69,11 +69,13 @@ class SinusoidalTimeEmbedding(nn.Module):
 class TemporalGRUGenerator(nn.Module):
     """Stage-1 temporal generator for sequence trend."""
 
-    def __init__(self, input_dim: int, hidden_dim: int = 256, num_layers: int = 1, dropout: float = 0.0):
+    def __init__(self, input_dim: int, hidden_dim: int = 256, num_layers: int = 1, dropout: float = 0.0, cond_dim: int = 0):
         super().__init__()
+        self.input_dim = int(input_dim)
+        self.cond_dim = int(cond_dim)
         self.start_token = nn.Parameter(torch.zeros(input_dim))
         self.gru = nn.GRU(
-            input_dim,
+            input_dim + self.cond_dim,
             hidden_dim,
             num_layers=num_layers,
             dropout=dropout if num_layers > 1 else 0.0,
@@ -81,11 +83,16 @@ class TemporalGRUGenerator(nn.Module):
         )
         self.out = nn.Linear(hidden_dim, input_dim)
 
-    def forward_teacher(self, x: torch.Tensor) -> torch.Tensor:
+    def forward_teacher(self, x: torch.Tensor, cond_cont: torch.Tensor | None = None) -> torch.Tensor:
         """Teacher-forced next-step prediction. Returns trend sequence and preds."""
         if x.size(1) < 2:
             raise ValueError("sequence length must be >= 2 for teacher forcing")
-        inp = x[:, :-1, :]
+        if self.cond_dim > 0:
+            if cond_cont is None:
+                cond_cont = torch.zeros(x.size(0), x.size(1), self.cond_dim, device=x.device, dtype=x.dtype)
+            inp = torch.cat([x[:, :-1, :], cond_cont[:, :-1, :]], dim=-1)
+        else:
+            inp = x[:, :-1, :]
         out, _ = self.gru(inp)
         pred_next = self.out(out)
         trend = torch.zeros_like(x)
@@ -93,13 +100,35 @@ class TemporalGRUGenerator(nn.Module):
         trend[:, 1:, :] = pred_next
         return trend, pred_next
 
-    def generate(self, batch_size: int, seq_len: int, device: torch.device) -> torch.Tensor:
+    def generate(
+        self,
+        batch_size: int,
+        seq_len: int,
+        device: torch.device,
+        cond_cont: torch.Tensor | None = None,
+        start_x: torch.Tensor | None = None,
+    ) -> torch.Tensor:
         """Autoregressively generate a backbone sequence."""
         h = None
-        prev = self.start_token.unsqueeze(0).expand(batch_size, -1).to(device)
+        if start_x is not None:
+            if start_x.dim() == 3 and start_x.size(1) == 1:
+                start_x = start_x[:, 0, :]
+            prev = start_x.to(device)
+        else:
+            prev = self.start_token.unsqueeze(0).expand(batch_size, -1).to(device)
+        if self.cond_dim > 0:
+            if cond_cont is None:
+                cond_cont = torch.zeros(batch_size, seq_len, self.cond_dim, device=device, dtype=prev.dtype)
+            else:
+                cond_cont = cond_cont.to(device)
         outputs = []
-        for _ in range(seq_len):
-            out, h = self.gru(prev.unsqueeze(1), h)
+        for t in range(seq_len):
+            if self.cond_dim > 0:
+                ct = cond_cont[:, t, :] if t < cond_cont.size(1) else torch.zeros(batch_size, self.cond_dim, device=device, dtype=prev.dtype)
+                step_inp = torch.cat([prev, ct], dim=-1)
+            else:
+                step_inp = prev
+            out, h = self.gru(step_inp.unsqueeze(1), h)
             nxt = self.out(out.squeeze(1))
             outputs.append(nxt.unsqueeze(1))
             prev = nxt
@@ -117,10 +146,13 @@ class TemporalTransformerGenerator(nn.Module):
         dropout: float = 0.1,
         pos_dim: int = 64,
         use_pos_embed: bool = True,
+        cond_dim: int = 0,
     ):
         super().__init__()
+        self.input_dim = int(input_dim)
+        self.cond_dim = int(cond_dim)
         self.start_token = nn.Parameter(torch.zeros(input_dim))
-        self.in_proj = nn.Linear(input_dim, hidden_dim)
+        self.in_proj = nn.Linear(input_dim + self.cond_dim, hidden_dim)
         self.pos_dim = pos_dim
         self.use_pos_embed = use_pos_embed
         self.pos_proj = nn.Linear(pos_dim, hidden_dim) if use_pos_embed and pos_dim > 0 else None
@@ -135,10 +167,15 @@ class TemporalTransformerGenerator(nn.Module):
         self.backbone = nn.TransformerEncoder(encoder_layer, num_layers=num_layers)
         self.out = nn.Linear(hidden_dim, input_dim)
 
-    def forward_teacher(self, x: torch.Tensor) -> torch.Tensor:
+    def forward_teacher(self, x: torch.Tensor, cond_cont: torch.Tensor | None = None) -> torch.Tensor:
         if x.size(1) < 2:
             raise ValueError("sequence length must be >= 2 for teacher forcing")
-        inp = x[:, :-1, :]
+        if self.cond_dim > 0:
+            if cond_cont is None:
+                cond_cont = torch.zeros(x.size(0), x.size(1), self.cond_dim, device=x.device, dtype=x.dtype)
+            inp = torch.cat([x[:, :-1, :], cond_cont[:, :-1, :]], dim=-1)
+        else:
+            inp = x[:, :-1, :]
         out = self._encode(inp)
         pred_next = self.out(out)
         trend = torch.zeros_like(x)
@@ -146,14 +183,40 @@ class TemporalTransformerGenerator(nn.Module):
         trend[:, 1:, :] = pred_next
         return trend, pred_next
 
-    def generate(self, batch_size: int, seq_len: int, device: torch.device) -> torch.Tensor:
-        context = self.start_token.unsqueeze(0).unsqueeze(1).expand(batch_size, 1, -1).to(device)
+    def generate(
+        self,
+        batch_size: int,
+        seq_len: int,
+        device: torch.device,
+        cond_cont: torch.Tensor | None = None,
+        start_x: torch.Tensor | None = None,
+    ) -> torch.Tensor:
+        if start_x is not None:
+            if start_x.dim() == 2:
+                context_x = start_x.unsqueeze(1).to(device)
+            else:
+                context_x = start_x.to(device)
+        else:
+            context_x = self.start_token.unsqueeze(0).unsqueeze(1).expand(batch_size, 1, -1).to(device)
+        if self.cond_dim > 0:
+            if cond_cont is None:
+                cond_cont = torch.zeros(batch_size, seq_len, self.cond_dim, device=device, dtype=context_x.dtype)
+            else:
+                cond_cont = cond_cont.to(device)
+                if cond_cont.size(1) < seq_len:
+                    pad = torch.zeros(batch_size, seq_len - cond_cont.size(1), self.cond_dim, device=device, dtype=cond_cont.dtype)
+                    cond_cont = torch.cat([cond_cont, pad], dim=1)
         outputs = []
         for _ in range(seq_len):
-            out = self._encode(context)
+            if self.cond_dim > 0:
+                cond_ctx = cond_cont[:, : context_x.size(1), :]
+                context_in = torch.cat([context_x, cond_ctx], dim=-1)
+            else:
+                context_in = context_x
+            out = self._encode(context_in)
             next_token = self.out(out[:, -1, :])
             outputs.append(next_token.unsqueeze(1))
-            context = torch.cat([context, next_token.unsqueeze(1)], dim=1)
+            context_x = torch.cat([context_x, next_token.unsqueeze(1)], dim=1)
         return torch.cat(outputs, dim=1)
 
     def _encode(self, x: torch.Tensor) -> torch.Tensor:

example/results/config_used.json

@@ -1,15 +1,15 @@
 {
-  "data_path": "/root/autodl-tmp/mask-ddpm/dataset/hai/hai-21.03/train1.csv.gz",
-  "data_glob": "/root/autodl-tmp/mask-ddpm/mask-ddpm/example/../../dataset/hai/hai-21.03/train*.csv.gz",
-  "split_path": "/root/autodl-tmp/mask-ddpm/mask-ddpm/example/feature_split.json",
-  "stats_path": "/root/autodl-tmp/mask-ddpm/mask-ddpm/example/results/cont_stats.json",
-  "vocab_path": "/root/autodl-tmp/mask-ddpm/mask-ddpm/example/results/disc_vocab.json",
-  "out_dir": "/root/autodl-tmp/mask-ddpm/mask-ddpm/example/results",
+  "data_path": "F:\\Development\\modbus_diffusion\\mask-ddpm\\dataset\\hai\\hai-21.03\\train1.csv.gz",
+  "data_glob": "F:\\Development\\modbus_diffusion\\mask-ddpm\\mask-ddpm\\example\\..\\..\\dataset\\hai\\hai-21.03\\train*.csv.gz",
+  "split_path": "F:\\Development\\modbus_diffusion\\mask-ddpm\\mask-ddpm\\example\\feature_split.json",
+  "stats_path": "F:\\Development\\modbus_diffusion\\mask-ddpm\\mask-ddpm\\example\\results\\cont_stats.json",
+  "vocab_path": "F:\\Development\\modbus_diffusion\\mask-ddpm\\mask-ddpm\\example\\results\\disc_vocab.json",
+  "out_dir": "F:\\Development\\modbus_diffusion\\mask-ddpm\\mask-ddpm\\example\\results",
   "device": "cuda",
   "timesteps": 600,
   "batch_size": 12,
   "seq_len": 96,
-  "epochs": 10,
+  "epochs": 0,
   "max_batches": 4000,
   "lambda": 0.7,
   "lr": 0.0005,
@@ -51,6 +51,8 @@
   "temporal_transformer_dropout": 0.1,
   "temporal_epochs": 3,
   "temporal_lr": 0.001,
+  "temporal_focus_type4": false,
+  "temporal_exclude_type4": false,
   "quantile_loss_weight": 0.2,
   "quantile_points": [
     0.05,
@@ -90,7 +92,11 @@
     "P1_PCV02Z",
     "P1_PCV01Z",
     "P1_PCV01D",
-    "P1_FCV02Z"
+    "P1_FCV02Z",
+    "P1_FCV03D",
+    "P1_FCV03Z",
+    "P1_LCV01D",
+    "P1_LCV01Z"
   ],
   "type4_features": [
     "P1_PIT02",
@@ -98,7 +104,8 @@
     "P1_FT03"
   ],
   "type5_features": [
-    "P1_FT03Z"
+    "P1_FT03Z",
+    "P1_FT02Z"
   ],
   "type6_features": [
     "P4_HT_PO",

example/results/train_log.csv

@@ -1,801 +1 @@
 epoch,step,loss,loss_cont,loss_disc
-0,0,3183.833496,4547.250000,2.165921
-0,10,4260.964844,6086.371094,1.477616
-0,20,2116.119873,3022.433350,1.257380
-0,30,3882.547119,5545.978516,1.078264
-0,40,2126.222656,3036.978516,0.966898
-0,50,2942.983643,4203.791504,0.981975
-0,60,3079.097900,4398.337402,0.751671
-0,70,2756.608154,3937.662598,0.690103
-0,80,2882.933350,4118.061523,0.833904
-0,90,3349.992188,4785.333496,0.757422
-0,100,1741.162476,2487.016113,0.741698
-0,110,1668.691528,2383.486084,0.736027
-0,120,1626.804077,2323.649658,0.714632
-0,130,1739.610596,2484.779053,0.805368
-0,140,1190.108276,1699.823975,0.678712
-0,150,1587.094727,2266.948242,0.682505
-0,160,1427.858887,2039.499756,0.609859
-0,170,1811.523193,2587.579590,0.641686
-0,180,1308.677856,1869.259033,0.509748
-0,190,1787.746948,2553.791260,0.202289
-0,200,1296.434937,1851.945190,0.162002
-0,210,1573.985596,2248.483643,0.062722
-0,220,1657.316772,2367.471191,0.214760
-0,230,1573.392700,2247.636719,0.068238
-0,240,1417.035889,2024.277466,0.049557
-0,250,1549.874023,2214.043213,0.047622
-0,260,1425.188721,2035.934326,0.038305
-0,270,1193.563599,1704.947876,0.255677
-0,280,1639.197632,2341.459473,0.510232
-0,290,1509.789917,2156.656494,0.358221
-0,300,1520.971924,2172.569580,0.499612
-0,310,1448.923096,2069.689209,0.405220
-0,320,1509.381592,2156.093506,0.324874
-0,330,1431.812378,2045.166016,0.590879
-0,340,1586.484375,2265.996826,0.895936
-0,350,1540.961792,2200.998291,0.807437
-0,360,1403.801270,2005.122559,0.650033
-0,370,1182.521973,1689.022095,0.607263
-0,380,1575.898926,2250.958496,0.672297
-0,390,1119.291748,1598.626587,0.761839
-0,400,1457.056763,2081.217529,0.623194
-0,410,1458.964355,2083.939697,0.627759
-0,420,1857.774902,2653.647217,0.657969
-0,430,1479.923218,2113.928467,0.519893
-0,440,1556.646851,2223.401611,0.798703
-0,450,1512.004028,2159.686768,0.684771
-0,460,1503.624146,2147.708740,0.707785
-0,470,1485.873047,2122.466064,0.437047
-0,480,1240.096558,1771.299438,0.569240
-0,490,1520.380493,2171.614014,0.783165
-0,500,1390.873169,1986.582886,0.827785
-0,510,1466.522095,2094.705078,0.713094
-0,520,1343.955078,1919.605347,0.707818
-0,530,1489.076904,2126.915771,0.734974
-0,540,1373.783813,1962.234985,0.680721
-0,550,1483.088501,2118.368652,0.715005
-0,560,1375.881714,1965.243042,0.646410
-0,570,1489.005127,2126.831543,0.696469
-0,580,1442.773315,2060.724854,0.844003
-0,590,1448.177979,2068.571533,0.545372
-0,600,1465.157715,2092.761230,0.707009
-0,610,1417.420410,2024.518066,0.810667
-0,620,1580.409302,2257.281982,0.981756
-0,630,1380.472656,1971.689453,0.917575
-0,640,1359.237793,1941.466919,0.650692
-0,650,1409.855713,2013.786499,0.632975
-0,660,1321.520020,1887.537476,0.760216
-0,670,1487.645386,2124.899902,0.669072
-0,680,1452.418335,2074.552979,0.725417
-0,690,1441.415161,2058.786133,0.834513
-0,700,1402.777588,2003.592529,0.817658
-0,710,1485.798706,2122.215820,0.770396
-0,720,1443.123169,2061.236084,0.801226
-0,730,1562.140747,2231.276367,0.760848
-0,740,1465.517090,2093.262695,0.718800
-0,750,1503.987549,2148.218994,0.731936
-0,760,1463.923584,2091.003906,0.680780
-0,770,1248.211670,1782.830566,0.706683
-0,780,1291.236084,1844.320435,0.651371
-0,790,1389.062012,1984.033691,0.709403
-1,0,1192.243164,1702.854980,0.694978
-1,10,1406.769165,2009.306519,0.767720
-1,20,1403.562134,2004.855957,0.472092
-1,30,1455.620605,2079.084473,0.809346
-1,40,1511.905029,2159.533203,0.690817
-1,50,1361.750732,1945.103149,0.523287
-1,60,1441.327515,2058.760498,0.602468
-1,70,1538.143433,2197.074707,0.577835
-1,80,1556.282959,2222.928955,0.707042
-1,90,1408.079956,2011.192383,0.755949
-1,100,1252.612061,1789.158936,0.587348
-1,110,1425.406250,2035.969116,0.702207
-1,120,1466.847290,2095.232422,0.559751
-1,130,1442.323730,2060.066650,0.864437
-1,140,1486.157593,2122.814209,0.568802
-1,150,1354.899292,1935.276489,0.622464
-1,160,1523.623169,2176.253174,0.764237
-1,170,1375.681519,1964.998901,0.545158
-1,180,1432.013916,2045.550781,0.364375
-1,190,1573.346191,2247.483154,0.293261
-1,200,1373.219238,1961.681885,0.073897
-1,210,1200.231689,1714.564941,0.044114
-1,220,1443.027466,2061.372070,0.153626
-1,230,1392.894165,1989.752197,0.159111
-1,240,1248.433350,1783.423218,0.045142
-1,250,1399.079956,1998.641724,0.046081
-1,260,1043.950562,1491.313354,0.026152
-1,270,1386.855713,1981.114624,0.190385
-1,280,1301.704956,1859.409668,0.327215
-1,290,1453.345703,2075.958252,0.534003
-1,300,1007.209778,1438.617432,0.507804
-1,310,1420.263184,2028.741821,0.432041
-1,320,1401.183838,2001.453613,0.500705
-1,330,1162.425659,1660.487305,0.229063
-1,340,1444.356934,2063.087402,0.597427
-1,350,1515.763794,2165.058105,0.693658
-1,360,1176.056152,1679.809082,0.579492
-1,370,1488.808594,2126.468262,0.875025
-1,380,1483.213379,2118.474854,0.880310
-1,390,1530.847290,2186.571533,0.770389
-1,400,1452.788818,2075.002686,0.902439
-1,410,1051.481812,1501.793091,0.699194
-1,420,1325.271729,1892.873901,0.817299
-1,430,1221.533203,1744.660034,0.826395
-1,440,1364.080688,1948.351685,0.644970
-1,450,1425.246582,2035.773926,0.599805
-1,460,1392.415894,1988.808228,0.766945
-1,470,1398.283203,1997.180542,0.804351
-1,480,1357.390625,1938.774780,0.782836
-1,490,1385.345947,1978.764404,0.652330
-1,500,1436.315063,2051.581055,0.641387
-1,510,1381.300049,1972.979248,0.665288
-1,520,1531.555908,2187.629395,0.669540
-1,530,1423.214111,2032.781616,0.844305
-1,540,1489.749756,2127.918945,0.644862
-1,550,1448.497803,2068.940674,0.748062
-1,560,1345.417603,1921.640747,0.847624
-1,570,1427.397217,2038.823608,0.692119
-1,580,1386.693604,1980.625854,0.800293
-1,590,1386.573486,1980.441284,0.837725
-1,600,1465.146729,2092.761719,0.667743
-1,610,1482.564819,2117.677246,0.584602
-1,620,1376.572876,1966.167847,0.800428
-1,630,918.296814,1311.554321,0.632072
-1,640,1324.733643,1892.135132,0.743117
-1,650,1401.460938,2001.751953,0.737485
-1,660,1447.588745,2067.669678,0.687752
-1,670,1245.077515,1778.398438,0.618556
-1,680,1341.766846,1916.551758,0.543834
-1,690,1405.754150,2007.927124,0.623327
-1,700,1443.610107,2061.937988,0.797176
-1,710,1411.496216,2016.130127,0.640040
-1,720,1356.189331,1937.049805,0.798009
-1,730,1274.552490,1820.477051,0.670119
-1,740,1423.458984,2033.200562,0.676492
-1,750,1479.755005,2113.521973,0.918397
-1,760,1450.786987,2072.183105,0.813719
-1,770,1281.817139,1830.844849,0.705245
-1,780,1346.415405,1923.209106,0.519566
-1,790,1358.163574,1939.853638,0.828874
-2,0,1310.122192,1871.381714,0.418646
-2,10,1361.304565,1944.360352,0.766878
-2,20,1466.903442,2095.298584,0.591117
-2,30,1188.091553,1696.908447,0.785655
-2,40,1031.407104,1473.176514,0.537891
-2,50,1450.158569,2071.397949,0.552217
-2,60,1406.581543,2009.071777,0.715261
-2,70,1425.637939,2036.380127,0.512891
-2,80,1281.622070,1830.561768,0.705330
-2,90,1403.056030,2004.125000,0.509431
-2,100,921.419739,1315.966064,0.738051
-2,110,1476.578735,2109.053467,0.758207
-2,120,1449.954224,2071.107666,0.521591
-2,130,1636.102051,2337.042969,0.519529
-2,140,1452.944824,2075.289551,0.746595
-2,150,1463.096558,2089.815918,0.694976
-2,160,1442.698608,2060.724121,0.573986
-2,170,1278.251953,1825.785767,0.605055
-2,180,1307.614746,1867.812744,0.415971
-2,190,1470.441772,2100.503418,0.248311
-2,200,1396.237671,1994.438232,0.381042
-2,210,1387.538574,1982.151001,0.054129
-2,220,1439.064453,2055.743652,0.090782
-2,230,1205.754150,1722.462646,0.061438
-2,240,1364.279419,1948.926514,0.055147
-2,250,1439.678833,2056.550781,0.255331
-2,260,1400.453857,2000.613892,0.035497
-2,270,1358.423462,1940.484863,0.233596
-2,280,1438.205200,2054.382324,0.407459
-2,290,1228.822510,1755.127930,0.715387
-2,300,1432.869751,2046.666626,0.633316
-2,310,1407.755005,2010.844116,0.493981
-2,320,1333.771729,1905.215942,0.346601
-2,330,1399.125488,1998.428223,0.710479
-2,340,1320.554199,1886.152832,0.781575
-2,350,1441.380615,2058.753662,0.800905
-2,360,1284.714478,1835.115845,0.395085
-2,370,1482.997559,2118.269043,0.645539
-2,380,1493.872192,2133.795166,0.676303
-2,390,1374.022339,1962.591675,0.646625
-2,400,1350.922241,1929.632324,0.554588
-2,410,1353.784302,1933.590454,0.853464
-2,420,1225.463135,1750.319458,0.742598
-2,430,1374.721436,1963.573486,0.688309
-2,440,1363.357910,1947.348022,0.670548
-2,450,1327.532227,1896.116577,0.789792
-2,460,1122.100342,1602.577515,0.931781
-2,470,1462.956421,2089.627197,0.680613
-2,480,1028.083862,1468.319336,0.826784
-2,490,1444.081421,2062.705322,0.585638
-2,500,1326.344604,1894.331787,0.993218
-2,510,1400.913696,2000.963989,0.748535
-2,520,1399.426270,1998.923096,0.551655
-2,530,1317.556152,1881.988647,0.506432
-2,540,1462.125244,2088.394531,0.785299
-2,550,1318.631836,1883.374390,0.853133
-2,560,1400.045166,1999.746948,0.703460
-2,570,1338.887451,1912.402100,0.640872
-2,580,1398.414673,1997.384155,0.776960
-2,590,1335.390625,1907.374634,0.717682
-2,600,1407.163574,2009.879639,0.782327
-2,610,1103.324097,1575.834595,0.737488
-2,620,1469.277954,2098.603271,0.803315
-2,630,1377.150635,1967.023315,0.728316
-2,640,1284.598755,1834.763550,0.835004
-2,650,1468.784790,2097.991455,0.595406
-2,660,1310.691284,1872.121460,0.631713
-2,670,1328.395996,1897.360840,0.761301
-2,680,1329.978882,1899.589111,0.844821
-2,690,1358.114746,1939.866577,0.648494
-2,700,1330.349365,1900.161133,0.732546
-2,710,1094.310059,1562.884155,0.919431
-2,720,996.794189,1423.642334,0.752403
-2,730,1509.052979,2155.485352,0.665574
-2,740,1437.262085,2052.928711,0.656775
-2,750,1192.932983,1703.909546,0.612598
-2,760,1379.242432,1969.965454,0.834590
-2,770,1392.717285,1989.165649,0.952917
-2,780,1409.844727,2013.728760,0.735940
-2,790,1354.926025,1935.198608,0.906308
-3,0,1474.789917,2106.657959,0.320913
-3,10,1233.481934,1761.912109,0.392319
-3,20,1306.513062,1866.142334,0.642290
-3,30,1382.109497,1974.100830,0.742490
-3,40,1445.045044,2063.961670,0.848702
-3,50,1220.986694,1743.925171,0.746243
-3,60,1394.458252,1991.725220,0.764863
-3,70,1016.398865,1451.727417,0.572467
-3,80,1347.343140,1924.485474,0.622466
-3,90,1479.106567,2112.760010,0.535122
-3,100,1501.767090,2145.130859,0.539186
-3,110,1398.487793,1997.606934,0.498223
-3,120,1387.138550,1981.401611,0.481850
-3,130,1333.497559,1904.718262,0.600926
-3,140,1406.692749,2009.258911,0.649231
-3,150,1528.006348,2182.471680,0.858538
-3,160,1194.264404,1705.728760,0.790984
-3,170,1344.019165,1919.763550,0.560959
-3,180,1385.327637,1978.867432,0.341166
-3,190,1249.843018,1785.375000,0.201357
-3,200,1339.098267,1912.930420,0.108677
-3,210,1351.928833,1931.249634,0.125045
-3,220,1321.489258,1887.758179,0.145731
-3,230,1481.554321,2116.449707,0.080655
-3,240,1444.743042,2063.873535,0.057254
-3,250,1311.291382,1873.229736,0.063893
-3,260,1436.432129,2052.005859,0.049077
-3,270,1325.572388,1893.491333,0.382297
-3,280,1379.625122,1970.720947,0.355222
-3,290,1244.760986,1777.933716,0.640860
-3,300,1251.521851,1787.569702,0.691775
-3,310,1193.661743,1704.999512,0.483678
-3,320,1392.481445,1989.121948,0.277586
-3,330,1371.599854,1959.125244,0.661989
-3,340,1388.293945,1983.006958,0.576749
-3,350,1410.410889,2014.557007,0.690085
-3,360,1290.459351,1843.239502,0.588218
-3,370,1386.642700,1980.658325,0.561619
-3,380,1431.838501,2045.134766,0.765934
-3,390,1330.940796,1900.982300,0.790070
-3,400,1386.725098,1980.703491,0.723482
-3,410,1281.702271,1830.694702,0.678544
-3,420,1450.241821,2071.481689,0.639323
-3,430,1334.888794,1906.645142,0.746211
-3,440,1475.799194,2107.986328,0.650364
-3,450,1324.678589,1892.029785,0.807135
-3,460,1354.497681,1934.654175,0.756229
-3,470,1107.910522,1582.490601,0.500695
-3,480,1435.326904,2050.153564,0.674719
-3,490,1319.159912,1884.216187,0.647862
-3,500,1340.236450,1914.410034,0.445445
-3,510,1342.714111,1917.880615,0.612413
-3,520,1348.723022,1926.403320,0.760435
-3,530,1276.793091,1823.782227,0.438722
-3,540,1405.454834,2007.481689,0.681101
-3,550,1302.241211,1859.974121,0.819137
-3,560,1539.127075,2198.369385,0.854926
-3,570,1331.934326,1902.481323,0.614280
-3,580,1194.027344,1705.411621,0.751001
-3,590,1362.927124,1946.679688,0.793889
-3,600,1361.725342,1944.994141,0.720772
-3,610,1373.232910,1961.463135,0.653718
-3,620,1302.502075,1860.420044,0.657502
-3,630,1317.537964,1881.926880,0.589975
-3,640,1428.997559,2041.073120,0.772419
-3,650,1255.266235,1792.909302,0.719631
-3,660,1291.690918,1844.971436,0.659025
-3,670,1378.569824,1969.075317,0.679549
-3,680,1198.053833,1711.111084,0.873652
-3,690,1250.473877,1786.015625,0.839820
-3,700,1291.293457,1844.336548,0.809987
-3,710,1389.156494,1984.216309,0.640415
-3,720,1451.439941,2073.153809,0.734372
-3,730,1389.186523,1984.201538,0.775694
-3,740,1460.354492,2085.836182,0.858047
-3,750,1494.142822,2134.183350,0.665175
-3,760,1398.958862,1998.135132,0.834417
-3,770,1360.851929,1943.689453,0.848515
-3,780,1288.482056,1840.310547,0.838317
-3,790,1344.724854,1920.667847,0.813874
-4,0,1331.611694,1902.077393,0.427302
-4,10,1351.088379,1929.804443,0.674919
-4,20,1344.562988,1920.547607,0.537479
-4,30,1380.292236,1971.577881,0.562771
-4,40,1476.475220,2108.975098,0.584272
-4,50,1437.855591,2053.761719,0.683065
-4,60,1447.503418,2067.565674,0.635268
-4,70,1396.583374,1994.818115,0.656298
-4,80,1402.370972,2003.057617,0.716676
-4,90,1463.434814,2090.305176,0.693182
-4,100,1411.198120,2015.713135,0.622315
-4,110,1334.692871,1906.357178,0.763460
-4,120,1379.127930,1969.912598,0.586174
-4,130,1254.672119,1792.081055,0.657634
-4,140,1264.629639,1806.285645,0.714038
-4,150,1438.762817,2055.045410,0.714952
-4,160,1412.637085,2017.642944,0.910181
-4,170,1364.711670,1949.279053,0.664152
-4,180,1377.968384,1968.219849,0.666259
-4,190,1494.332520,2134.637451,0.229824
-4,200,1355.711548,1936.567505,0.320178
-4,210,1419.343994,2027.532837,0.179927
-4,220,1143.623535,1633.558594,0.391724
-4,230,1454.817993,2078.259033,0.068267
-4,240,1330.723755,1900.961304,0.117085
-4,250,1415.292480,2021.797607,0.066861
-4,260,1369.316406,1956.117920,0.069667
-4,270,1443.334351,2061.570801,0.715960
-4,280,1381.048340,1972.641846,0.615013
-4,290,1400.020752,1999.707031,0.697925
-4,300,1270.261475,1814.306152,0.769272
-4,310,1416.783691,2023.819092,0.320928
-4,320,1380.434082,1971.820557,0.481759
-4,330,1394.524292,1991.854858,0.708896
-4,340,1390.549316,1986.302368,0.411659
-4,350,1219.830078,1742.383301,0.492701
-4,360,1394.687866,1992.046143,0.802289
-4,370,1384.779663,1977.907593,0.764099
-4,380,1425.712646,2036.517334,0.454959
-4,390,1269.325195,1812.961182,0.793848
-4,400,1332.023926,1902.674072,0.465408
-4,410,1358.205078,1940.052124,0.525438
-4,420,1218.857666,1740.845581,0.846466
-4,430,1432.824097,2046.588867,0.657311
-4,440,1397.947754,1996.774902,0.645416
-4,450,1317.465576,1881.677002,0.935239
-4,460,1437.938110,2053.898926,0.657603
-4,470,1333.029053,1903.984863,0.765121
-4,480,1379.205811,1970.075195,0.460714
-4,490,1403.384155,2004.584229,0.541554
-4,500,1500.303101,2143.006592,0.618541
-4,510,1469.422119,2098.944092,0.497694
-4,520,1407.687744,2010.706177,0.604236
-4,530,1324.212891,1891.329956,0.901385
-4,540,1051.848145,1502.319458,0.692303
-4,550,1410.138550,2014.073853,0.910204
-4,560,1080.340576,1543.012939,0.707045
-4,570,1480.853760,2115.173828,0.736971
-4,580,1313.857056,1876.673950,0.572590
-4,590,1489.191772,2127.010742,0.908589
-4,600,1402.813232,2003.766235,0.545628
-4,610,1350.005127,1928.306152,0.596006
-4,620,1409.174561,2012.693848,0.919607
-4,630,1322.290894,1888.672974,0.685984
-4,640,1483.182739,2118.432373,0.887685
-4,650,1477.207397,2109.938721,0.795490
-4,660,1368.469482,1954.636963,0.703189
-4,670,1346.890259,1923.822388,0.679141
-4,680,1309.446167,1870.284790,0.782634
-4,690,1259.538818,1798.951660,0.871608
-4,700,1213.964722,1733.989868,0.538498
-4,710,1341.846313,1916.611084,0.688046
-4,720,1278.633911,1826.321167,0.655422
-4,730,1321.792969,1888.046021,0.496103
-4,740,1294.553955,1849.028320,0.734913
-4,750,1331.466309,1901.765991,0.728413
-4,760,1343.263062,1918.590332,0.793747
-4,770,1346.412720,1923.192139,0.549143
-4,780,1394.250732,1991.488403,0.654786
-4,790,1178.614258,1683.442993,0.624984
-5,0,1381.578125,1973.440308,0.475622
-5,10,1292.605347,1846.350342,0.445175
-5,20,1343.879395,1919.609863,0.449178
-5,30,1333.156250,1904.241333,0.568427
-5,40,1432.652100,2046.270264,0.826029
-5,50,1424.269287,2034.384644,0.615697
-5,60,1386.381226,1980.285889,0.559561
-5,70,1296.574829,1852.006958,0.525181
-5,80,1235.729736,1765.068481,0.552107
-5,90,1388.295532,1983.003906,0.598889
-5,100,1362.903809,1946.667480,0.742882
-5,110,1339.192017,1912.866089,0.579080
-5,120,1445.760620,2065.082031,0.623679
-5,130,1444.509766,2063.282715,0.659761
-5,140,1260.166138,1799.897583,0.722035
-5,150,1311.296509,1872.888184,0.857408
-5,160,1401.315552,2001.600098,0.595261
-5,170,1431.771118,2045.029053,0.781770
-5,180,1476.872559,2109.530518,0.611725
-5,190,1433.121338,2047.075928,0.497609
-5,200,1368.606201,1954.957397,0.398229
-5,210,1434.444092,2049.137451,0.108338
-5,220,1301.695435,1859.457764,0.198909
-5,230,1350.648560,1929.392822,0.198544
-5,240,1402.938965,2004.110107,0.158466
-5,250,1374.445557,1963.446411,0.063809
-5,260,1331.590454,1902.227417,0.058766
-5,270,1358.835938,1940.958740,0.490266
-5,280,1277.334961,1824.541626,0.471479
-5,290,1289.189087,1841.444946,0.543997
-5,300,1234.718262,1763.685425,0.398413
-5,310,1467.128052,2095.673828,0.481397
-5,320,1484.625122,2120.722412,0.351158
-5,330,1402.348145,2003.108276,0.524456
-5,340,1311.093872,1872.732666,0.560579
-5,350,1299.273682,1855.947510,0.321461
-5,360,1318.420410,1883.184204,0.597162
-5,370,1356.793457,1937.954956,0.708372
-5,380,1215.855225,1736.743164,0.393075
-5,390,1391.972534,1988.185425,0.768275
-5,400,1385.025635,1978.317627,0.634894
-5,410,1338.120117,1911.285156,0.697956
-5,420,1053.971558,1505.272339,0.894339
-5,430,1466.043457,2094.085938,0.571332
-5,440,1446.653687,2066.309326,0.749992
-5,450,1352.591431,1931.964844,0.680090
-5,460,938.970459,1341.028442,0.775030
-5,470,1433.025024,2046.961548,0.462366
-5,480,1382.997559,1975.474731,0.502687
-5,490,1211.074097,1729.823120,0.616560
-5,500,1355.821167,1936.559326,0.721231
-5,510,1220.898071,1743.921021,0.469647
-5,520,1311.034302,1872.702148,0.435444
-5,530,1475.334595,2107.316162,0.661365
-5,540,1202.873535,1718.044434,0.772177
-5,550,1368.494263,1954.595703,0.884758
-5,560,1352.476685,1931.809448,0.660819
-5,570,1265.553101,1807.529419,0.899061
-5,580,1359.037720,1941.129883,0.781809
-5,590,1433.665527,2047.720947,0.830307
-5,600,1454.692383,2077.746826,0.855085
-5,610,1350.079468,1928.400513,0.622484
-5,620,1431.371338,2044.401855,0.924907
-5,630,1353.085205,1932.621704,0.789309
-5,640,1343.684692,1919.240234,0.680895
-5,650,1153.452271,1647.436646,0.782024
-5,660,1341.494995,1916.067017,0.784926
-5,670,1206.219971,1722.879272,0.635128
-5,680,1368.555542,1954.773193,0.669750
-5,690,1199.484131,1713.189453,0.800724
-5,700,1394.796143,1992.213379,0.783516
-5,710,1293.515015,1847.551025,0.719907
-5,720,1390.605835,1986.249878,0.720971
-5,730,1321.877930,1888.095459,0.664969
-5,740,1449.051514,2069.736572,0.746680
-5,750,1337.472046,1910.338745,0.741736
-5,760,1256.858887,1795.156982,0.784702
-5,770,1306.135498,1865.594971,0.684015
-5,780,1315.913208,1879.570679,0.670323
-5,790,1451.889648,2073.807861,0.705214
-6,0,1402.109131,2002.675049,0.703659
-6,10,1414.470215,2020.408325,0.529152
-6,20,1377.476929,1967.583374,0.494593
-6,30,1421.947144,2031.060059,0.627377
-6,40,1337.385376,1910.196289,0.772749
-6,50,1331.247559,1901.492065,0.629588
-6,60,1363.626343,1947.782349,0.544360
-6,70,1306.166992,1865.672729,0.607935
-6,80,1394.703735,1992.135986,0.647641
-6,90,1295.145264,1849.924438,0.615009
-6,100,1306.985474,1866.828979,0.640840
-6,110,1469.330078,2098.781006,0.566757
-6,120,1292.434814,1846.058960,0.596419
-6,130,1090.012207,1556.880127,0.602326
-6,140,1271.001099,1815.373901,0.744468
-6,150,1371.640137,1959.215820,0.581206
-6,160,1352.763550,1932.159424,0.777695
-6,170,1335.919189,1908.079102,0.824870
-6,180,1324.261230,1891.578613,0.454496
-6,190,1302.863525,1860.899902,0.708667
-6,200,1193.828003,1705.381592,0.129727
-6,210,1375.309448,1964.645874,0.131988
-6,220,1289.626587,1842.238892,0.142794
-6,230,1382.268066,1974.597290,0.117433
-6,240,1396.389526,1994.690674,0.304747
-6,250,1212.410278,1731.955566,0.082068
-6,260,1442.449951,2060.592285,0.071983
-6,270,1349.172485,1927.286011,0.195212
-6,280,1267.510132,1810.568237,0.312127
-6,290,1458.786133,2083.804443,0.361726
-6,300,1258.974976,1798.295044,0.513440
-6,310,1461.175293,2087.189697,0.433528
-6,320,1230.999023,1758.362671,0.437663
-6,330,1386.516357,1980.402344,0.735239
-6,340,1442.661743,2060.596924,0.768169
-6,350,1309.266968,1870.097168,0.614782
-6,360,1420.350708,2028.842529,0.485295
-6,370,1412.714355,2017.947144,0.460662
-6,380,1426.125366,2037.005371,0.693541
-6,390,1321.096069,1886.950073,0.721198
-6,400,1483.022705,2118.393311,0.445550
-6,410,1365.421875,1950.244751,0.783437
-6,420,1436.181030,2051.374023,0.686178
-6,430,1482.874023,2118.070068,0.712547
-6,440,1302.881104,1861.025269,0.493659
-6,450,1356.820068,1937.925049,0.866001
-6,460,1446.284668,2065.759033,0.801515
-6,470,1034.633789,1477.762329,0.611554
-6,480,1263.159912,1804.263428,0.546072
-6,490,1448.827148,2069.524414,0.490979
-6,500,1365.781860,1950.902466,0.458113
-6,510,1392.418579,1988.891724,0.603457
-6,520,1380.563232,1971.895264,0.748071
-6,530,1383.598877,1976.245605,0.714526
-6,540,1283.676392,1833.432251,0.870318
-6,550,1406.994019,2009.698486,0.642495
-6,560,1452.090210,2074.053711,0.800106
-6,570,1128.043213,1611.086792,0.888306
-6,580,1353.439575,1933.157593,0.716758
-6,590,1433.040283,2046.892822,0.678801
-6,600,1465.141479,2092.778564,0.611183
-6,610,1360.478882,1943.135986,0.905106
-6,620,1365.959229,1951.064087,0.674060
-6,630,1364.727905,1949.238647,0.827151
-6,640,1276.682983,1823.518188,0.687770
-6,650,1314.865479,1878.089233,0.635976
-6,660,1448.328125,2068.691406,0.775176
-6,670,1424.035522,2033.954468,0.852288
-6,680,1385.819824,1979.389648,0.779933
-6,690,1443.240845,2061.336426,0.976319
-6,700,1326.837524,1895.168579,0.680883
-6,710,1408.063965,2011.209473,0.689842
-6,720,1393.958618,1991.055786,0.683554
-6,730,1383.852539,1976.616577,0.698341
-6,740,1371.577515,1959.052368,0.763584
-6,750,975.784180,1393.620850,0.774587
-6,760,1397.252808,1995.782959,0.639068
-6,770,1296.468140,1851.828003,0.580253
-6,780,1323.732544,1890.660278,0.853858
-6,790,1253.175171,1789.927856,0.707006
-7,0,1427.383911,2038.921875,0.357666
-7,10,1149.285278,1641.530762,0.625638
-7,20,1398.872803,1998.107422,0.595522
-7,30,1173.637939,1676.378174,0.510373
-7,40,1408.580688,2011.977295,0.602553
-7,50,1327.318359,1895.835449,0.726305
-7,60,1210.938477,1729.561646,0.769907
-7,70,1289.211426,1841.411621,0.701483
-7,80,1256.855591,1795.246094,0.573556
-7,90,1404.395386,2005.986816,0.637843
-7,100,1333.393188,1904.515625,0.731328
-7,110,1380.505615,1971.832520,0.699024
-7,120,1124.031982,1605.477051,0.587771
-7,130,1247.642334,1782.053345,0.637281
-7,140,1414.689087,2020.663818,0.699980
-7,150,1467.120850,2095.564209,0.702573
-7,160,1394.851196,1992.263428,0.830261
-7,170,1479.298584,2112.988037,0.628267
-7,180,1391.184082,1987.135620,0.573001
-7,190,1573.935059,2248.292236,0.379493
-7,200,1334.128540,1905.747681,0.291988
-7,210,1410.637695,2015.101440,0.172436
-7,220,1369.421509,1956.244873,0.117897
-7,230,1397.929321,1996.967407,0.129837
-7,240,1275.251709,1821.724243,0.100066
-7,250,1277.143677,1824.428589,0.091271
-7,260,1397.119019,1995.833862,0.070850
-7,270,1262.207031,1803.056519,0.179170
-7,280,1394.497925,1991.944824,0.405396
-7,290,1329.105835,1898.488892,0.496955
-7,300,1440.054810,2056.928955,0.632547
-7,310,1296.870728,1852.476440,0.413340
-7,320,1375.024780,1964.122925,0.416442
-7,330,1339.952637,1914.008179,0.442784
-7,340,1552.520752,2217.615479,0.592493
-7,350,1338.742798,1912.194458,0.642815
-7,360,1396.040405,1994.139893,0.427340
-7,370,1206.552368,1723.403931,0.523065
-7,380,1217.284180,1738.697998,0.603473
-7,390,1465.359985,2093.057617,0.687396
-7,400,1223.174194,1746.990356,0.894657
-7,410,1322.689087,1889.255981,0.647148
-7,420,1426.747925,2037.823730,0.863030
-7,430,1327.786865,1896.571411,0.578874
-7,440,1400.174561,1999.928711,0.705682
-7,450,1393.601929,1990.545532,0.691348
-7,460,1225.392700,1750.205078,0.791842
-7,470,1377.230713,1967.231079,0.515584
-7,480,993.232178,1418.604248,0.640837
-7,490,1385.767822,1979.260498,0.907338
-7,500,1370.890503,1958.112183,0.657997
-7,510,1337.819458,1910.889038,0.615062
-7,520,1370.637695,1957.712280,0.748476
-7,530,1338.053833,1911.143799,0.798569
-7,540,1366.044678,1951.188599,0.670380
-7,550,1393.078125,1989.796753,0.700469
-7,560,1129.454224,1613.093140,0.911564
-7,570,1273.085571,1818.362671,0.726649
-7,580,1265.287842,1807.219116,0.737303
-7,590,1386.473145,1980.346313,0.729028
-7,600,1316.534424,1880.404053,0.802621
-7,610,1462.629395,2089.238770,0.500190
-7,620,1343.828369,1919.334839,0.936861
-7,630,1185.687134,1693.479736,0.792609
-7,640,1329.250122,1898.628174,0.661207
-7,650,1451.651245,2073.404297,0.852252
-7,660,1341.429077,1916.085693,0.529053
-7,670,1357.372314,1938.728271,0.833373
-7,680,1288.235962,1839.983154,0.784858
-7,690,1364.239868,1948.524292,0.869382
-7,700,1322.240845,1888.505005,0.923278
-7,710,1377.459351,1967.501099,0.643483
-7,720,1246.564819,1780.465942,0.750393
-7,730,1425.477295,2036.080566,0.692054
-7,740,1351.905151,1931.035278,0.557867
-7,750,1388.324341,1982.980103,0.746014
-7,760,1384.515015,1977.564697,0.687298
-7,770,1158.278320,1654.387329,0.634209
-7,780,1348.044312,1925.471069,0.679789
-7,790,1184.838013,1692.351685,0.594770
-8,0,1430.853882,2043.857178,0.413180
-8,10,1402.583740,2003.463989,0.447673
-8,20,1354.354248,1934.396606,0.859101
-8,30,1412.740845,2017.922974,0.591015
-8,40,1305.915283,1865.248901,0.736410
-8,50,1276.975342,1823.904053,0.757966
-8,60,1384.281860,1977.309692,0.492115
-8,70,1267.346558,1810.141724,0.783241
-8,80,1345.061768,1921.269043,0.537711
-8,90,1346.264404,1922.974854,0.563284
-8,100,1391.877808,1988.105957,0.635657
-8,110,1366.934448,1952.426147,0.741894
-8,120,1253.876343,1790.927246,0.701687
-8,130,1277.523926,1824.749512,0.613267
-8,140,1303.589233,1861.918335,0.766769
-8,150,1379.761719,1970.743286,0.749061
-8,160,1256.161743,1794.232178,0.607267
-8,170,1286.510132,1837.542114,0.709287
-8,180,1381.922485,1973.952271,0.458755
-8,190,1367.694824,1953.644531,0.422003
-8,200,1351.914429,1931.200439,0.190074
-8,210,1310.535278,1872.065308,0.242793
-8,220,1074.070679,1534.313232,0.108120
-8,230,1422.666748,2032.280640,0.182237
-8,240,1348.740234,1926.643921,0.252413
-8,250,1394.232544,1991.698730,0.102536
-8,260,1382.515137,1974.969849,0.076682
-8,270,1407.324463,2010.302490,0.327881
-8,280,1362.257446,1945.829346,0.531096
-8,290,1386.497925,1980.541992,0.351584
-8,300,1460.178955,2085.657715,0.683763
-8,310,1275.496216,1821.935913,0.424736
-8,320,1385.472656,1979.039185,0.430447
-8,330,1517.117188,2167.053955,0.548888
-8,340,1318.596558,1883.497803,0.452265
-8,350,1242.619263,1774.961182,0.440472
-8,360,1310.204834,1871.395752,0.715310
-8,370,1401.934570,2002.489258,0.599136
-8,380,1193.227905,1704.293823,0.692336
-8,390,1329.633667,1899.126221,0.774669
-8,400,1350.765991,1929.382446,0.614092
-8,410,1405.248291,2007.186035,0.683590
-8,420,1396.550781,1994.799561,0.597365
-8,430,1364.058960,1948.376343,0.609830
-8,440,1409.152954,2012.791016,0.621008
-8,450,1306.067871,1865.509888,0.652660
-8,460,1374.333984,1962.987915,0.770055
-8,470,1306.677246,1866.374268,0.672749
-8,480,1365.465210,1950.428833,0.503327
-8,490,1369.738525,1956.447144,0.707809
-8,500,1446.645508,2066.331055,0.672423
-8,510,1404.481201,2006.176270,0.475139
-8,520,1333.953003,1905.416016,0.499908
-8,530,1451.401978,2073.205322,0.485541
-8,540,1276.721802,1823.650391,0.513795
-8,550,1416.303223,2022.925537,0.801875
-8,560,1512.302246,2160.107910,0.715139
-8,570,1458.925903,2083.821777,0.796904
-8,580,1489.223511,2127.107666,0.781716
-8,590,1340.004639,1914.004272,0.623736
-8,600,1354.776978,1934.992798,0.899712
-8,610,1225.583008,1750.554321,0.603478
-8,620,1225.543213,1750.441528,0.743612
-8,630,1356.140137,1937.092529,0.543510
-8,640,1357.819824,1939.369995,0.830926
-8,650,1276.639404,1823.498413,0.595685
-8,660,1352.599243,1931.996704,0.628689
-8,670,1326.531128,1894.691162,0.778533
-8,680,1168.036743,1668.287476,0.735241
-8,690,1183.291138,1690.091187,0.707204
-8,700,1406.184814,2008.546631,0.638747
-8,710,1462.970459,2089.560791,0.880348
-8,720,1411.845337,2016.594727,0.727399
-8,730,1183.481201,1690.309448,0.845188
-8,740,1375.187744,1964.222900,0.728351
-8,750,1382.059326,1973.986206,0.855727
-8,760,1352.645508,1932.000977,0.773716
-8,770,1240.372070,1771.634155,0.712623
-8,780,1351.737671,1930.695068,0.793972
-8,790,1362.313965,1945.845337,0.701692
-9,0,1167.117920,1667.099609,0.374608
-9,10,1389.198242,1984.362549,0.404507
-9,20,1284.041260,1834.031250,0.669447
-9,30,1257.346313,1795.982422,0.462772
-9,40,1342.695557,1917.816772,0.689729
-9,50,1360.405273,1943.096558,0.743577
-9,60,1301.845825,1859.481934,0.645122
-9,70,1485.945679,2122.514648,0.570123
-9,80,1461.847900,2088.071045,0.618146
-9,90,1314.727783,1877.804321,0.840805
-9,100,1334.130127,1905.605103,0.645316
-9,110,1221.221191,1744.278931,0.699416
-9,120,1385.631104,1979.259644,0.450594
-9,130,1327.047119,1895.480713,0.654597
-9,140,1381.455688,1973.164307,0.751353
-9,150,1213.203613,1732.836060,0.663023
-9,160,1238.085083,1768.382080,0.664160
-9,170,1262.798096,1803.687988,0.660173
-9,180,1430.566772,2043.460205,0.422453
-9,190,1270.846558,1815.304443,0.378631
-9,200,1422.160645,2031.538330,0.223065
-9,210,1374.248169,1963.124878,0.147017
-9,220,1298.134521,1854.404053,0.120393
-9,230,1349.836914,1928.244385,0.168487
-9,240,1354.287598,1934.576172,0.234960
-9,250,1452.162231,2074.443604,0.133361
-9,260,1409.717163,2013.821289,0.097083
-9,270,1420.777954,2029.524048,0.323000
-9,280,1397.178223,1995.813965,0.314937
-9,290,1382.241455,1974.427612,0.425842
-9,300,1360.468384,1943.337036,0.394178
-9,310,1291.504395,1844.836426,0.349840
-9,320,1405.881104,2008.145874,0.556430
-9,330,1257.617676,1796.228394,0.815588
-9,340,1255.381226,1793.084961,0.690902
-9,350,1288.353394,1840.301392,0.418188
-9,360,1246.255249,1780.006470,0.781109
-9,370,1322.656616,1889.198242,0.680039
-9,380,1295.011841,1849.683105,0.737371
-9,390,1400.574951,2000.593140,0.485168
-9,400,1426.532471,2037.596191,0.678564
-9,410,1378.598389,1969.005981,0.938101
-9,420,1282.483398,1831.788696,0.721778
-9,430,1258.534546,1797.566162,0.753230
-9,440,1386.989014,1981.104004,0.677300
-9,450,1401.886353,2002.432007,0.570592
-9,460,1422.267090,2031.503296,0.677374
-9,470,1360.145752,1942.751953,0.695174
-9,480,1416.858032,2023.820068,0.569829
-9,490,1353.596069,1933.342407,0.804701
-9,500,1363.734009,1947.892456,0.662382
-9,510,1357.725220,1939.232300,0.834168
-9,520,1362.017456,1945.438477,0.662265
-9,530,1236.691650,1766.357422,0.755286
-9,540,1447.441772,2067.534668,0.516392
-9,550,1312.263550,1874.350586,0.682969
-9,560,1303.780762,1862.239014,0.664007
-9,570,1446.870361,2066.586670,0.819120
-9,580,1427.273560,2038.615967,0.766797
-9,590,1398.765869,1997.956299,0.615517
-9,600,1372.102295,1959.864990,0.612538
-9,610,1351.152100,1929.911865,0.671775
-9,620,1339.440674,1913.146851,0.750727
-9,630,1256.030273,1793.962769,0.809595
-9,640,1431.364136,2044.424072,0.851141
-9,650,1197.544678,1710.423950,0.785345
-9,660,1364.932739,1949.630493,0.600471
-9,670,1202.534668,1717.592529,0.687238
-9,680,1413.439819,2018.868896,0.729740
-9,690,1297.329346,1852.953979,0.827475
-9,700,1394.275146,1991.509033,0.685360
-9,710,1329.347412,1898.767334,0.656253
-9,720,1231.160889,1758.465088,0.729312
-9,730,1374.985352,1963.928467,0.739654
-9,740,1223.265869,1747.197144,0.711846
-9,750,1398.930420,1998.148315,0.716038
-9,760,1281.642944,1830.601807,0.698900
-9,770,1328.708618,1897.798340,0.789865
-9,780,1293.412109,1847.345093,0.847453
-9,790,1441.486572,2058.928955,0.744339

example/results/type1_debug.json

@@ -0,0 +1,38 @@
+{
+  "features": [
+    "P1_B4022",
+    "P4_HT_LD"
+  ],
+  "generated": {
+    "P1_B4022": {
+      "num_changes": 383,
+      "mean_dwell": 1.0,
+      "median_dwell": 1.0,
+      "mean_step": 0.01383368146214119,
+      "median_step": 0.00916000000000139
+    },
+    "P4_HT_LD": {
+      "num_changes": 15,
+      "mean_dwell": 24.0,
+      "median_dwell": 2.5,
+      "mean_step": 0.03375866666666667,
+      "median_step": 0.02171
+    }
+  },
+  "reference": {
+    "P1_B4022": {
+      "num_changes": 909932,
+      "mean_dwell": 1.0128251200912595,
+      "median_dwell": 1,
+      "mean_step": 0.01171526248115434,
+      "median_step": 0.00952999999999804
+    },
+    "P4_HT_LD": {
+      "num_changes": 585364,
+      "mean_dwell": 1.5744074210108223,
+      "median_dwell": 1,
+      "mean_step": 1.043229230819962,
+      "median_step": 0.24593999999999738
+    }
+  }
+}

example/train.py

@@ -79,6 +79,8 @@ DEFAULTS = {
     "temporal_transformer_dropout": 0.1,
     "temporal_epochs": 2,
     "temporal_lr": 1e-3,
+    "temporal_focus_type4": False,
+    "temporal_exclude_type4": False,
     "quantile_loss_weight": 0.0,
     "quantile_points": [0.05, 0.25, 0.5, 0.75, 0.95],
     "snr_weighted_loss": True,
@@ -110,6 +112,7 @@ def parse_args():
     parser.add_argument("--device", default="auto", help="cpu, cuda, or auto")
     parser.add_argument("--out-dir", default=None, help="Override output directory")
     parser.add_argument("--seed", type=int, default=None, help="Override random seed")
+    parser.add_argument("--temporal-only", action="store_true", help="Only train temporal stage-1 and exit.")
     return parser.parse_args()
 
 
@@ -178,6 +181,9 @@ def main():
         config["out_dir"] = str(out_dir)
     if args.seed is not None:
         config["seed"] = int(args.seed)
+    if bool(args.temporal_only):
+        config["use_temporal_stage1"] = True
+        config["epochs"] = 0
 
     set_seed(int(config["seed"]))
 
@@ -188,8 +194,10 @@ def main():
 
     type1_cols = config.get("type1_features", []) or []
     type5_cols = config.get("type5_features", []) or []
+    type4_cols = config.get("type4_features", []) or []
     type1_cols = [c for c in type1_cols if c in cont_cols]
     type5_cols = [c for c in type5_cols if c in cont_cols]
+    type4_cols = [c for c in type4_cols if c in cont_cols]
     model_cont_cols = [c for c in cont_cols if c not in type1_cols and c not in type5_cols]
     if not model_cont_cols:
         raise SystemExit("model_cont_cols is empty; check type1/type5 config")
@@ -243,6 +251,18 @@ def main():
     opt = torch.optim.Adam(model.parameters(), lr=float(config["lr"]))
     temporal_model = None
     opt_temporal = None
+    temporal_use_type1_cond = bool(config.get("temporal_use_type1_cond", False))
+    temporal_cond_dim = len(type1_cols) if (temporal_use_type1_cond and type1_cols) else 0
+    temporal_focus_type4 = bool(config.get("temporal_focus_type4", False))
+    temporal_exclude_type4 = bool(config.get("temporal_exclude_type4", False))
+    type4_model_idx = [model_cont_cols.index(c) for c in type4_cols if c in model_cont_cols]
+    trend_mask = None
+    if temporal_focus_type4 and type4_model_idx:
+        trend_mask = torch.zeros(1, 1, len(model_cont_cols), device=device)
+        trend_mask[:, :, type4_model_idx] = 1.0
+    elif temporal_exclude_type4 and type4_model_idx:
+        trend_mask = torch.ones(1, 1, len(model_cont_cols), device=device)
+        trend_mask[:, :, type4_model_idx] = 0.0
     if bool(config.get("use_temporal_stage1", False)):
         temporal_backbone = str(config.get("temporal_backbone", "gru"))
         if temporal_backbone == "transformer":
@@ -255,6 +275,7 @@ def main():
                 dropout=float(config.get("temporal_transformer_dropout", 0.1)),
                 pos_dim=int(config.get("temporal_pos_dim", 64)),
                 use_pos_embed=bool(config.get("temporal_use_pos_embed", True)),
+                cond_dim=temporal_cond_dim,
             ).to(device)
         else:
             temporal_model = TemporalGRUGenerator(
@@ -262,6 +283,7 @@ def main():
                 hidden_dim=int(config.get("temporal_hidden_dim", 256)),
                 num_layers=int(config.get("temporal_num_layers", 1)),
                 dropout=float(config.get("temporal_dropout", 0.0)),
+                cond_dim=temporal_cond_dim,
             ).to(device)
         opt_temporal = torch.optim.Adam(
             temporal_model.parameters(),
@@ -306,8 +328,18 @@ def main():
                 x_cont = x_cont.to(device)
                 model_idx = [cont_cols.index(c) for c in model_cont_cols]
                 x_cont_model = x_cont[:, :, model_idx]
-                trend, pred_next = temporal_model.forward_teacher(x_cont_model)
-                temporal_loss = F.mse_loss(pred_next, x_cont_model[:, 1:, :])
+                cond_cont = None
+                if temporal_cond_dim > 0:
+                    cond_idx = [cont_cols.index(c) for c in type1_cols]
+                    cond_cont = x_cont[:, :, cond_idx]
+                trend, pred_next = temporal_model.forward_teacher(x_cont_model, cond_cont=cond_cont)
+                target_next = x_cont_model[:, 1:, :]
+                if trend_mask is not None:
+                    mask = trend_mask.to(dtype=pred_next.dtype, device=pred_next.device)
+                    mse = (pred_next - target_next) ** 2
+                    temporal_loss = (mse * mask).sum() / torch.clamp(mask.sum() * mse.size(0) * mse.size(1), min=1.0)
+                else:
+                    temporal_loss = F.mse_loss(pred_next, target_next)
                 opt_temporal.zero_grad()
                 temporal_loss.backward()
                 if float(config.get("grad_clip", 0.0)) > 0:
@@ -356,7 +388,9 @@ def main():
             if temporal_model is not None:
                 temporal_model.eval()
                 with torch.no_grad():
-                    trend, _ = temporal_model.forward_teacher(x_cont_model)
+                    trend, _ = temporal_model.forward_teacher(x_cont_model, cond_cont=cond_cont)
+                    if trend_mask is not None and trend is not None:
+                        trend = trend * trend_mask.to(dtype=trend.dtype, device=trend.device)
             x_cont_resid = x_cont_model if trend is None else x_cont_model - trend
 
             bsz = x_cont.size(0)