first commit

models/ETSformer.py

@@ -0,0 +1,110 @@
+import torch
+import torch.nn as nn
+from layers.Embed import DataEmbedding
+from layers.ETSformer_EncDec import EncoderLayer, Encoder, DecoderLayer, Decoder, Transform
+
+
+class Model(nn.Module):
+    """
+    Paper link: https://arxiv.org/abs/2202.01381
+    """
+
+    def __init__(self, configs):
+        super(Model, self).__init__()
+        self.task_name = configs.task_name
+        self.seq_len = configs.seq_len
+        self.label_len = configs.label_len
+        if self.task_name == 'classification' or self.task_name == 'anomaly_detection' or self.task_name == 'imputation':
+            self.pred_len = configs.seq_len
+        else:
+            self.pred_len = configs.pred_len
+
+        assert configs.e_layers == configs.d_layers, "Encoder and decoder layers must be equal"
+
+        # Embedding
+        self.enc_embedding = DataEmbedding(configs.enc_in, configs.d_model, configs.embed, configs.freq,
+                                           configs.dropout)
+
+        # Encoder
+        self.encoder = Encoder(
+            [
+                EncoderLayer(
+                    configs.d_model, configs.n_heads, configs.enc_in, configs.seq_len, self.pred_len, configs.top_k,
+                    dim_feedforward=configs.d_ff,
+                    dropout=configs.dropout,
+                    activation=configs.activation,
+                ) for _ in range(configs.e_layers)
+            ]
+        )
+        # Decoder
+        self.decoder = Decoder(
+            [
+                DecoderLayer(
+                    configs.d_model, configs.n_heads, configs.c_out, self.pred_len,
+                    dropout=configs.dropout,
+                ) for _ in range(configs.d_layers)
+            ],
+        )
+        self.transform = Transform(sigma=0.2)
+
+        if self.task_name == 'classification':
+            self.act = torch.nn.functional.gelu
+            self.dropout = nn.Dropout(configs.dropout)
+            self.projection = nn.Linear(configs.d_model * configs.seq_len, configs.num_class)
+
+    def forecast(self, x_enc, x_mark_enc, x_dec, x_mark_dec):
+        with torch.no_grad():
+            if self.training:
+                x_enc = self.transform.transform(x_enc)
+        res = self.enc_embedding(x_enc, x_mark_enc)
+        level, growths, seasons = self.encoder(res, x_enc, attn_mask=None)
+
+        growth, season = self.decoder(growths, seasons)
+        preds = level[:, -1:] + growth + season
+        return preds
+
+    def imputation(self, x_enc, x_mark_enc, x_dec, x_mark_dec, mask):
+        res = self.enc_embedding(x_enc, x_mark_enc)
+        level, growths, seasons = self.encoder(res, x_enc, attn_mask=None)
+        growth, season = self.decoder(growths, seasons)
+        preds = level[:, -1:] + growth + season
+        return preds
+
+    def anomaly_detection(self, x_enc):
+        res = self.enc_embedding(x_enc, None)
+        level, growths, seasons = self.encoder(res, x_enc, attn_mask=None)
+        growth, season = self.decoder(growths, seasons)
+        preds = level[:, -1:] + growth + season
+        return preds
+
+    def classification(self, x_enc, x_mark_enc):
+        res = self.enc_embedding(x_enc, None)
+        _, growths, seasons = self.encoder(res, x_enc, attn_mask=None)
+
+        growths = torch.sum(torch.stack(growths, 0), 0)[:, :self.seq_len, :]
+        seasons = torch.sum(torch.stack(seasons, 0), 0)[:, :self.seq_len, :]
+
+        enc_out = growths + seasons
+        output = self.act(enc_out)  # the output transformer encoder/decoder embeddings don't include non-linearity
+        output = self.dropout(output)
+
+        # Output
+        output = output * x_mark_enc.unsqueeze(-1)  # zero-out padding embeddings
+        output = output.reshape(output.shape[0], -1)  # (batch_size, seq_length * d_model)
+        output = self.projection(output)  # (batch_size, num_classes)
+        return output
+
+    def forward(self, x_enc, x_mark_enc, x_dec, x_mark_dec, mask=None):
+        if self.task_name == 'long_term_forecast' or self.task_name == 'short_term_forecast':
+            dec_out = self.forecast(x_enc, x_mark_enc, x_dec, x_mark_dec)
+            return dec_out[:, -self.pred_len:, :]  # [B, L, D]
+        if self.task_name == 'imputation':
+            dec_out = self.imputation(x_enc, x_mark_enc, x_dec, x_mark_dec, mask)
+            return dec_out  # [B, L, D]
+        if self.task_name == 'anomaly_detection':
+            dec_out = self.anomaly_detection(x_enc)
+            return dec_out  # [B, L, D]
+        if self.task_name == 'classification':
+            dec_out = self.classification(x_enc, x_mark_enc)
+            return dec_out  # [B, N]
+        return None