fix model

pengsihua2023 · Nov 4, 2024 · 4ff2bc2 · 4ff2bc2
1 parent 6b77591
commit 4ff2bc2
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 38 deletions.
diff --git a/models/Timer.py b/models/Timer.py
@@ -41,39 +41,6 @@ def __init__(self, configs):
                 else:
                     raise NotImplementedError
 
-    def encoder_top(self, x_enc, x_mark_enc, x_dec, x_mark_dec):
-        # Normalization from Non-stationary Transformer
-        means = x_enc.mean(1, keepdim=True).detach()
-        x_enc = x_enc - means
-        stdev = torch.sqrt(torch.var(x_enc, dim=1, keepdim=True, unbiased=False) + 1e-5).detach()
-        x_enc /= stdev
-
-        # do patching and embedding
-        x_enc = x_enc.permute(0, 2, 1)
-        # u: [bs * nvars x patch_num x d_model]
-        dec_in, n_vars = self.enc_embedding(x_enc)
-
-        # Encoder
-        # z: [bs * nvars x patch_num x d_model]
-
-        return dec_in
-
-    def encoder_bottom(self, x_enc, x_mark_enc, x_dec, x_mark_dec):
-        # Normalization from Non-stationary Transformer
-        means = x_enc.mean(1, keepdim=True).detach()
-        x_enc = x_enc - means
-        stdev = torch.sqrt(torch.var(x_enc, dim=1, keepdim=True, unbiased=False) + 1e-5).detach()
-        x_enc /= stdev
-
-        # do patching and embedding
-        x_enc = x_enc.permute(0, 2, 1)
-        # u: [bs * nvars x patch_num x d_model]
-        dec_in, n_vars = self.enc_embedding(x_enc) # [B * M, N, D]
-
-        # Encoder
-        dec_out, attns = self.decoder(dec_in) # [B * M, N, D]
-        return dec_out
-
     def forecast(self, x_enc, x_mark_enc, x_dec, x_mark_dec, mask=None):
         B, L, M = x_enc.shape
 
@@ -87,7 +54,7 @@ def forecast(self, x_enc, x_mark_enc, x_dec, x_mark_dec, mask=None):
         x_enc = x_enc.permute(0, 2, 1) # [B, M, T]
         dec_in, n_vars = self.enc_embedding(x_enc) # [B * M, N, D]
 
-        # Encoder
+        # Transformer Blocks
         dec_out, attns = self.decoder(dec_in) # [B * M, N, D]
         dec_out = self.proj(dec_out) # [B * M, N, L]
         dec_out = dec_out.reshape(B, M, -1).transpose(1, 2) # [B, T, M]
@@ -114,7 +81,7 @@ def imputation(self, x_enc, x_mark_enc, x_dec, x_mark_dec, mask=None):
         x_enc = x_enc.permute(0, 2, 1) # [B, M, T]
         dec_in, n_vars = self.enc_embedding(x_enc) # [B * M, N, D]
 
-        # Encoder
+        # Transformer Blocks
         dec_out, attns = self.decoder(dec_in) # [B * M, N, D]
         dec_out = self.proj(dec_out) # [B * M, N, L]
         dec_out = dec_out.reshape(B, M, -1).transpose(1, 2) # [B, T, M]
@@ -136,7 +103,7 @@ def anomaly_detection(self, x_enc):
         x_enc = x_enc.permute(0, 2, 1) # [B, M, T]
         dec_in, n_vars = self.enc_embedding(x_enc) # [B * M, N, D]
 
-        # Encoder
+        # Transformer Blocks
         dec_out, attns = self.decoder(dec_in) # [B * M, N, D]
         dec_out = self.proj(dec_out) # [B * M, N, L]
         dec_out = dec_out.reshape(B, M, -1).transpose(1, 2) # [B, T, M]

diff --git a/models/TimerBackbone.py b/models/TimerBackbone.py
@@ -23,10 +23,10 @@ def __init__(self, configs):
         self.patch_embedding = PatchEmbedding(
             self.d_model, self.patch_len, self.stride, padding, self.dropout)
 
-        # Decoder
+        # Decoder-only Transformer: Refer to issue: https://github.com/thuml/Large-Time-Series-Model/issues/23
         self.decoder = Encoder(
             [
-                EncoderLayer(
+                 EncoderLayer(
                     AttentionLayer(
                         FullAttention(True, configs.factor, attention_dropout=configs.dropout,
                                       output_attention=True), configs.d_model, configs.n_heads),