config/ASTRA/BJTaxi-InFlow.yaml

@@ -32,6 +32,7 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  output_dim: 1
 
 train:
   batch_size: 32

config/ASTRA/BJTaxi-OutFlow.yaml

@@ -17,6 +17,7 @@ data:
   steps_per_day: 48
   test_ratio: 0.2
   val_ratio: 0.2
+  output_dim: 1
   
 model:
   d_ff: 128

config/ASTRA/METR-LA.yaml

@@ -32,6 +32,7 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  output_dim: 1
 
 train:
   batch_size: 16

config/ASTRA/NYCBike-InFlow.yaml

@@ -32,6 +32,7 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  output_dim: 1
   
 train:
   batch_size: 32

config/ASTRA/NYCBike-OutFlow.yaml

@@ -32,6 +32,7 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  output_dim: 1
 
 train:
   batch_size: 32

config/ASTRA/PEMS-BAY.yaml

@@ -32,6 +32,7 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  output_dim: 1
 
 train:
   batch_size: 16

config/ASTRA/SolarEnergy.yaml

@@ -32,6 +32,7 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  output_dim: 1
 
 train:
   batch_size: 64

config/ASTRA_v2/AirQuality.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -33,6 +34,8 @@ model:
   stride: 7
   word_num: 1000
   output_dim: 6
+  graph_dim: 64
+  graph_embed_dim: 10
 
 train:
   batch_size: 16

config/ASTRA_v2/BJTaxi-InFlow.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 32

config/ASTRA_v2/BJTaxi-OutFlow.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 32

config/ASTRA_v2/METR-LA.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 16

config/ASTRA_v2/NYCBike-InFlow.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 32

config/ASTRA_v2/NYCBike-OutFlow.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 32

config/ASTRA_v2/PEMS-BAY.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 16

config/ASTRA_v2/SolarEnergy.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 16

config/ASTRA_v3/AirQuality.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -33,6 +34,9 @@ model:
   stride: 7
   word_num: 1000
   output_dim: 6
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 6
 
 train:
   batch_size: 16

config/ASTRA_v3/BJTaxi-InFlow.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 32

config/ASTRA_v3/BJTaxi-OutFlow.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 32

config/ASTRA_v3/METR-LA.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 16

config/ASTRA_v3/NYCBike-InFlow.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 32

config/ASTRA_v3/NYCBike-OutFlow.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 32

config/ASTRA_v3/PEMS-BAY.yaml

@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,6 +33,9 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
   batch_size: 16

config/ASTRA_v3/SolarEnergy.yaml

@@ -6,7 +6,7 @@ basic:
   seed: 2023
 
 data:
-  batch_size: 64
+  batch_size: 16
   column_wise: false
   days_per_week: 7
   horizon: 24
@@ -19,6 +19,7 @@ data:
   val_ratio: 0.2
 
 model:
+  cheb: 3
   d_ff: 128
   d_model: 64
   dropout: 0.2
@@ -32,9 +33,12 @@ model:
   seq_len: 24
   stride: 7
   word_num: 1000
+  graph_dim: 64
+  graph_embed_dim: 10
+  output_dim: 1
 
 train:
-  batch_size: 64
+  batch_size: 16
   debug: false
   early_stop: true
   early_stop_patience: 15

model/ASTRA/astrav2.py

@@ -127,8 +127,11 @@ class ASTRA(nn.Module):
         self.gpt_layers = configs['gpt_layers']  # 使用的GPT2层数
         self.d_ff = configs['d_ff']  # 前馈网络隐藏层维度
         self.gpt_path = configs['gpt_path']  # 预训练GPT2模型路径
-        self.num_nodes = configs.get('num_nodes', 325)  # 节点数量
-        self.output_dim = configs.get('output_dim', 1)
+        self.num_nodes = configs['num_nodes']  # 节点数量
+        self.output_dim = configs['output_dim']
+        self.cheb = configs['cheb']
+        self.graph_dim = configs['graph_dim']
+        self.graph_embed_dim = configs['graph_embed_dim']
 
         self.word_choice = GumbelSoftmax(configs['word_num'])  # 词汇选择层
 
@@ -152,18 +155,18 @@ class ASTRA(nn.Module):
         
         # 初始化图增强编码器
         self.graph_encoder = GraphEnhancedEncoder(
-            K=configs.get('chebyshev_order', 3),  # Chebyshev多项式阶数
+            K=self.cheb,  # Chebyshev多项式阶数
             in_dim=self.d_model,  # 输入特征维度
-            hidden_dim=configs.get('graph_hidden_dim', 32),  # 隐藏层维度
+            hidden_dim=self.graph_dim,  # 隐藏层维度
             num_nodes=self.num_nodes,  # 节点数量
-            embed_dim=configs.get('graph_embed_dim', 10),  # 节点嵌入维度
+            embed_dim=self.graph_embed_dim,  # 节点嵌入维度
             device=self.device,  # 运行设备
             temporal_dim=self.seq_len,  # 时间序列长度
             num_features=self.input_dim  # 特征通道数
         )
         
         self.graph_projection = nn.Linear(  # 图特征投影层，每一k阶的切比雪夫权重映射到隐藏维度
-            configs.get('graph_hidden_dim', 32) * (configs.get('chebyshev_order', 3) + 1),  # 输入维度
+            self.graph_dim * (self.cheb + 1),  # 输入维度
             self.d_model  # 输出维度
         )
     

model/ASTRA/astrav3.py

@@ -127,8 +127,11 @@ class ASTRA(nn.Module):
         self.gpt_layers = configs['gpt_layers']  # 使用的GPT2层数
         self.d_ff = configs['d_ff']  # 前馈网络隐藏层维度
         self.gpt_path = configs['gpt_path']  # 预训练GPT2模型路径
-        self.num_nodes = configs.get('num_nodes', 325)  # 节点数量
-        self.output_dim = configs.get('output_dim', 1)
+        self.num_nodes = configs['num_nodes'] # 节点数量
+        self.output_dim = configs['output_dim']
+        self.cheb = configs['cheb']
+        self.graph_dim = configs['graph_dim']
+        self.graph_embed_dim = configs['graph_embed_dim']
 
         self.word_choice = GumbelSoftmax(configs['word_num'])  # 词汇选择层
 
@@ -148,22 +151,22 @@ class ASTRA(nn.Module):
         self.word_embeddings = self.gpts.get_input_embeddings().weight.to(self.device)  # 词嵌入权重
         self.vocab_size = self.word_embeddings.shape[0]  # 词汇表大小
         self.mapping_layer = nn.Linear(self.vocab_size, 1)  # 映射层
-        self.reprogramming_layer = ReprogrammingLayer(self.d_model + configs.get('graph_hidden_dim', 32) * (configs.get('chebyshev_order', 3) + 1), self.n_heads, self.d_keys, self.d_llm)  # 重编程层
+        self.reprogramming_layer = ReprogrammingLayer(self.d_model + self.graph_dim * (self.cheb + 1), self.n_heads, self.d_keys, self.d_llm)  # 重编程层
         
         # 初始化图增强编码器
         self.graph_encoder = GraphEnhancedEncoder(
             K=configs.get('chebyshev_order', 3),  # Chebyshev多项式阶数
             in_dim=self.d_model,  # 输入特征维度
-            hidden_dim=configs.get('graph_hidden_dim', 32),  # 隐藏层维度
+            hidden_dim=self.graph_dim,  # 隐藏层维度
             num_nodes=self.num_nodes,  # 节点数量
-            embed_dim=configs.get('graph_embed_dim', 10),  # 节点嵌入维度
+            embed_dim=self.graph_embed_dim,  # 节点嵌入维度
             device=self.device,  # 运行设备
             temporal_dim=self.seq_len,  # 时间序列长度
             num_features=self.input_dim  # 特征通道数
         )
         
         self.graph_projection = nn.Linear(  # 图特征投影层，每一k阶的切比雪夫权重映射到隐藏维度
-            configs.get('graph_hidden_dim', 32) * (configs.get('chebyshev_order', 3) + 1),  # 输入维度
+            self.graph_dim * (self.cheb + 1),  # 输入维度
             self.d_model # 输出维度
         )
     

model/REPST/repst.py

@@ -19,7 +19,7 @@ class repst(nn.Module):
         self.gpt_layers = configs['gpt_layers']
         self.d_ff = configs['d_ff']
         self.gpt_path = configs['gpt_path']
-        self.output_dim = configs.get('output_dim', 1)
+        self.output_dim = configs['output_dim']
 
         self.word_choice = GumbelSoftmax(configs['word_num'])
 

train.py

@@ -90,7 +90,7 @@ def main(model, data, debug=False):
 if __name__ == "__main__":
     # 调试用
     # model_list = ["iTransformer", "PatchTST", "HI"]
-    model_list = ["ASTRA_v3", "ASTRA_v2", "ASTRA", "REPST", "STAEFormer", "MTGNN", "iTransformer", "PatchTST", "HI"]
+    model_list = ["ASTRA_v3"]
     # model_list = ["MTGNN"]
     # dataset_list = ["AirQuality", "SolarEnergy", "PEMS-BAY", "METR-LA", "BJTaxi-InFlow", "BJTaxi-OutFlow", "NYCBike-InFlow", "NYCBike-OutFlow"]
     # dataset_list = ["AirQuality"]