import torch import torch.nn as nn import torch.nn.functional as F import torch.fft import numpy as np # Basit embedding ve conv blocks - layers klasörü olmadan class DataEmbedding(nn.Module): def __init__(self, c_in, d_model, embed_type='fixed', freq='h', dropout=0.1, seq_len=6000): super(DataEmbedding, self).__init__() self.c_in = c_in self.d_model = d_model self.embed_type = embed_type self.freq = freq self.seq_len = seq_len # Basit linear embedding self.value_embedding = nn.Linear(c_in, d_model) # Position embedding'i seq_len'e göre oluştur self.position_embedding = nn.Parameter(torch.randn(1, seq_len, d_model)) self.dropout = nn.Dropout(p=dropout) def forward(self, x, x_mark): x = self.value_embedding(x) # Position embedding'i input boyutuna göre crop et # seq_len'e göre oluşturulduğu için genelde uyumlu olacak if x.size(1) <= self.position_embedding.size(1): x = x + self.position_embedding[:, :x.size(1), :] else: # Eğer input daha büyükse, position embedding'i extend et x = x + self.position_embedding remaining_length = x.size(1) - self.position_embedding.size(1) if remaining_length > 0: # Sinusoidal position encoding ekle pos_encoding = self._get_sinusoidal_encoding(remaining_length, self.d_model) pos_encoding = pos_encoding.unsqueeze(0).to(x.device) x[:, self.position_embedding.size(1):, :] += pos_encoding return self.dropout(x) def _get_sinusoidal_encoding(self, length, d_model): """Sinusoidal position encoding oluştur""" position = torch.arange(length).unsqueeze(1).float() div_term = torch.exp(torch.arange(0, d_model, 2).float() * -(np.log(10000.0) / d_model)) pos_encoding = torch.zeros(length, d_model) pos_encoding[:, 0::2] = torch.sin(position * div_term) pos_encoding[:, 1::2] = torch.cos(position * div_term) return pos_encoding class Inception_Block_V1(nn.Module): def __init__(self, in_channels, out_channels, num_kernels=6, init_weight=True): super(Inception_Block_V1, self).__init__() self.in_channels = in_channels self.out_channels = out_channels self.num_kernels = num_kernels kernels = [] for i in range(self.num_kernels): kernels.append(nn.Conv2d(in_channels, out_channels, kernel_size=2 * i + 1, padding=i)) self.kernels = nn.ModuleList(kernels) if init_weight: self._initialize_weights() def _initialize_weights(self): for m in self.modules(): if isinstance(m, nn.Conv2d): nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu') if m.bias is not None: nn.init.constant_(m.bias, 0) def forward(self, x): res_list = [] for i, kernel in enumerate(self.kernels): res_list.append(kernel(x)) res = torch.stack(res_list, dim=-1).mean(-1) return res def FFT_for_Period(x, k=2): # [B, T, C] xf = torch.fft.rfft(x, dim=1) # find period by amplitudes frequency_list = abs(xf).mean(0).mean(-1) frequency_list[0] = 0 _, top_list = torch.topk(frequency_list, k) top_list = top_list.detach().cpu().numpy() period = x.shape[1] // top_list return period, abs(xf).mean(-1)[:, top_list] class TimesBlock(nn.Module): def __init__(self, configs): super(TimesBlock, self).__init__() self.seq_len = configs.seq_len self.pred_len = configs.pred_len self.k = configs.top_k # parameter-efficient design self.conv = nn.Sequential( Inception_Block_V1(configs.d_model, configs.d_ff, num_kernels=configs.num_kernels), nn.GELU(), Inception_Block_V1(configs.d_ff, configs.d_model, num_kernels=configs.num_kernels) ) def forward(self, x): B, T, N = x.size() #B: batch size T: length of time series N:number of features period_list, period_weight = FFT_for_Period(x, self.k) res = [] for i in range(self.k): period = period_list[i] # padding if (self.seq_len + self.pred_len) % period != 0: length = ( ((self.seq_len + self.pred_len) // period) + 1) * period padding = torch.zeros([x.shape[0], (length - (self.seq_len + self.pred_len)), x.shape[2]]).to(x.device) out = torch.cat([x, padding], dim=1) else: length = (self.seq_len + self.pred_len) out = x # reshape out = out.reshape(B, length // period, period, N).permute(0, 3, 1, 2).contiguous() # 2D conv: from 1d Variation to 2d Variation out = self.conv(out) # reshape back out = out.permute(0, 2, 3, 1).reshape(B, -1, N) res.append(out[:, :(self.seq_len + self.pred_len), :]) res = torch.stack(res, dim=-1) # adaptive aggregation period_weight = F.softmax(period_weight, dim=1) period_weight = period_weight.unsqueeze( 1).unsqueeze(1).repeat(1, T, N, 1) res = torch.sum(res * period_weight, -1) # residual connection res = res + x return res class Model(nn.Module): """ Paper link: https://openreview.net/pdf?id=ju_Uqw384Oq """ def __init__(self, configs): super(Model, self).__init__() self.configs = configs self.task_name = configs.task_name self.seq_len = configs.seq_len self.label_len = configs.label_len self.pred_len = configs.pred_len self.model = nn.ModuleList([TimesBlock(configs) for _ in range(configs.e_layers)]) self.enc_embedding = DataEmbedding(configs.enc_in, configs.d_model, configs.embed, configs.freq, configs.dropout, configs.seq_len) self.layer = configs.e_layers self.layer_norm = nn.LayerNorm(configs.d_model) if self.task_name == 'long_term_forecast' or self.task_name == 'short_term_forecast': self.predict_linear = nn.Linear( self.seq_len, self.pred_len + self.seq_len) self.projection = nn.Linear( configs.d_model, configs.c_out, bias=True) if self.task_name == 'imputation' or self.task_name == 'anomaly_detection': self.projection = nn.Linear( configs.d_model, configs.c_out, bias=True) # Transfer learning için P-S prediction heads (sadece gerektiğinde eklenir) if hasattr(configs, 'use_ps_heads') and configs.use_ps_heads: # Skip attention for memory efficiency - use only pooling # Multi-scale feature extraction (reduced sizes for memory) self.multi_scale_pools = nn.ModuleList([ nn.AdaptiveAvgPool1d(16), # Local patterns (reduced) nn.AdaptiveAvgPool1d(4), # Medium patterns nn.AdaptiveAvgPool1d(1), # Global patterns ]) # Feature fusion - calculate exact dimension # Pool sizes: 16 + 4 + 1 = 21, so total dim = d_model * 21 fusion_dim = configs.d_model * (16 + 4 + 1) # Exact calculation self.feature_fusion = nn.Sequential( nn.Linear(fusion_dim, configs.d_model), nn.ReLU(), nn.Dropout(configs.dropout) ) # Separate P and S regression heads self.p_regression_head = nn.Sequential( nn.Linear(configs.d_model, 128), nn.ReLU(), nn.Dropout(configs.dropout), nn.Linear(128, 64), nn.ReLU(), nn.Dropout(configs.dropout), nn.Linear(64, 1) # P time only ) self.s_regression_head = nn.Sequential( nn.Linear(configs.d_model, 128), nn.ReLU(), nn.Dropout(configs.dropout), nn.Linear(128, 64), nn.ReLU(), nn.Dropout(configs.dropout), nn.Linear(64, 1) # S time only ) # Separate P and S classification heads self.p_classification_head = nn.Sequential( nn.Linear(configs.d_model, 64), nn.ReLU(), nn.Dropout(configs.dropout), nn.Linear(64, 32), nn.ReLU(), nn.Dropout(configs.dropout), nn.Linear(32, 1), # P exists/not nn.Sigmoid() ) self.s_classification_head = nn.Sequential( nn.Linear(configs.d_model, 64), nn.ReLU(), nn.Dropout(configs.dropout), nn.Linear(64, 32), nn.ReLU(), nn.Dropout(configs.dropout), nn.Linear(32, 1), # S exists/not nn.Sigmoid() ) if self.task_name == 'classification': self.act = F.gelu self.dropout = nn.Dropout(configs.dropout) self.projection = nn.Linear( configs.d_model * configs.seq_len, configs.num_class) def anomaly_detection(self, x_enc): # Transfer learning için P-S heads varsa - SADECE ONLARI KULLAN if hasattr(self, 'p_regression_head'): # 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) x_enc /= stdev # embedding enc_out = self.enc_embedding(x_enc, None) # [B,T,C] # TimesNet for i in range(self.layer): enc_out = self.layer_norm(self.model[i](enc_out)) # Skip attention for memory - use direct multi-scale pooling # Multi-scale feature extraction directly on TimesNet output enc_out_transposed = enc_out.permute(0, 2, 1) # (B, d_model, T) multi_scale_features = [] # Manual pooling for large sequences to avoid CUDA memory issues pool_sizes = [16, 4, 1] # Target pool sizes for i, target_size in enumerate(pool_sizes): T = enc_out_transposed.size(2) # Sequence length if T >= 8000: # Very large - use manual avg pooling # Manual average pooling window_size = T // target_size if window_size > 0: # Reshape and average # (B, d_model, T) -> (B, d_model, target_size, window_size) trimmed_T = (T // window_size) * window_size trimmed = enc_out_transposed[:, :, :trimmed_T] reshaped = trimmed.view(trimmed.size(0), trimmed.size(1), target_size, window_size) pooled = reshaped.mean(dim=3) # Average over window else: # Fallback: simple reshape pooled = enc_out_transposed[:, :, :target_size] if T >= target_size else enc_out_transposed else: # Use normal adaptive pooling for smaller sequences pool = self.multi_scale_pools[i] pooled = pool(enc_out_transposed) # (B, d_model, pool_size) flattened = pooled.flatten(1) # (B, d_model * pool_size) multi_scale_features.append(flattened) # Concatenate multi-scale features fused_features = torch.cat(multi_scale_features, dim=1) # (B, d_model * 3) # Feature fusion final_features = self.feature_fusion(fused_features) # (B, d_model) # Separate P and S predictions p_time = self.p_regression_head(final_features) # (B, 1) s_time = self.s_regression_head(final_features) # (B, 1) ps_times = torch.cat([p_time, s_time], dim=1) # (B, 2) # Separate P and S classifications p_class = self.p_classification_head(final_features) # (B, 1) s_class = self.s_classification_head(final_features) # (B, 1) ps_classification = torch.cat([p_class, s_class], dim=1) # (B, 2) return ps_times, ps_classification else: # Orijinal anomaly detection (reconstruction) # 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) x_enc /= stdev # embedding enc_out = self.enc_embedding(x_enc, None) # [B,T,C] # TimesNet for i in range(self.layer): enc_out = self.layer_norm(self.model[i](enc_out)) # porject back dec_out = self.projection(enc_out) # De-Normalization from Non-stationary Transformer dec_out = dec_out * \ (stdev[:, 0, :].unsqueeze(1).repeat( 1, self.pred_len + self.seq_len, 1)) dec_out = dec_out + \ (means[:, 0, :].unsqueeze(1).repeat( 1, self.pred_len + self.seq_len, 1)) return dec_out 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': result = self.anomaly_detection(x_enc) return result # [B, L, D] veya [B, L, D], [B, 2], [B, 1] if self.task_name == 'classification': dec_out = self.classification(x_enc, x_mark_enc) return dec_out # [B, N] return None def forecast(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) x_enc /= stdev # embedding enc_out = self.enc_embedding(x_enc, x_mark_enc) # [B,T,C] enc_out = self.predict_linear(enc_out.permute(0, 2, 1)).permute( 0, 2, 1) # align temporal dimension # TimesNet for i in range(self.layer): enc_out = self.layer_norm(self.model[i](enc_out)) # porject back dec_out = self.projection(enc_out) # De-Normalization from Non-stationary Transformer dec_out = dec_out * \ (stdev[:, 0, :].unsqueeze(1).repeat( 1, self.pred_len + self.seq_len, 1)) dec_out = dec_out + \ (means[:, 0, :].unsqueeze(1).repeat( 1, self.pred_len + self.seq_len, 1)) return dec_out def imputation(self, x_enc, x_mark_enc, x_dec, x_mark_dec, mask): # Normalization from Non-stationary Transformer means = torch.sum(x_enc, dim=1) / torch.sum(mask == 1, dim=1) means = means.unsqueeze(1).detach() x_enc = x_enc - means x_enc = x_enc.masked_fill(mask == 0, 0) stdev = torch.sqrt(torch.sum(x_enc * x_enc, dim=1) / torch.sum(mask == 1, dim=1) + 1e-5) stdev = stdev.unsqueeze(1).detach() x_enc /= stdev # embedding enc_out = self.enc_embedding(x_enc, x_mark_enc) # [B,T,C] # TimesNet for i in range(self.layer): enc_out = self.layer_norm(self.model[i](enc_out)) # porject back dec_out = self.projection(enc_out) # De-Normalization from Non-stationary Transformer dec_out = dec_out * \ (stdev[:, 0, :].unsqueeze(1).repeat( 1, self.pred_len + self.seq_len, 1)) dec_out = dec_out + \ (means[:, 0, :].unsqueeze(1).repeat( 1, self.pred_len + self.seq_len, 1)) return dec_out def classification(self, x_enc, x_mark_enc): # embedding enc_out = self.enc_embedding(x_enc, None) # [B,T,C] # TimesNet for i in range(self.layer): enc_out = self.layer_norm(self.model[i](enc_out)) # Output # the output transformer encoder/decoder embeddings don't include non-linearity output = self.act(enc_out) output = self.dropout(output) # zero-out padding embeddings output = output * x_mark_enc.unsqueeze(-1) # (batch_size, seq_length * d_model) output = output.reshape(output.shape[0], -1) output = self.projection(output) # (batch_size, num_classes) return output