dsipts 1.1.5__py3-none-any.whl

dsipts/models/patchtst/layers.py

@@ -0,0 +1,569 @@
+
+
+import torch
+from torch import nn
+import math
+from typing import Optional
+from torch import Tensor
+import torch.nn.functional as F
+import numpy as np
+
+
+class Transpose(nn.Module):
+    def __init__(self, *dims, contiguous=False): 
+        super().__init__()
+        self.dims, self.contiguous = dims, contiguous
+    def forward(self, x):
+        if self.contiguous: 
+            return x.transpose(*self.dims).contiguous()
+        else:
+            return x.transpose(*self.dims)
+
+    
+
+    
+# decomposition
+
+class moving_avg(nn.Module):
+    """
+    Moving average block to highlight the trend of time series
+    """
+    def __init__(self, kernel_size, stride):
+        super(moving_avg, self).__init__()
+        self.kernel_size = kernel_size
+        self.avg = nn.AvgPool1d(kernel_size=kernel_size, stride=stride, padding=0)
+
+    def forward(self, x):
+        # padding on the both ends of time series
+        front = x[:, 0:1, :].repeat(1, (self.kernel_size - 1) // 2, 1)
+        end = x[:, -1:, :].repeat(1, (self.kernel_size - 1) // 2, 1)
+        x = torch.cat([front, x, end], dim=1)
+        x = self.avg(x.permute(0, 2, 1))
+        x = x.permute(0, 2, 1)
+        return x
+
+
+class series_decomp(nn.Module):
+    """
+    Series decomposition block
+    """
+    def __init__(self, kernel_size):
+        super(series_decomp, self).__init__()
+        self.moving_avg = moving_avg(kernel_size, stride=1)
+
+    def forward(self, x):
+        moving_mean = self.moving_avg(x)
+        res = x - moving_mean
+        return res, moving_mean
+    
+    
+    
+# pos_encoding
+
+def PositionalEncoding(q_len, d_model, normalize=True):
+    pe = torch.zeros(q_len, d_model)
+    position = torch.arange(0, q_len).unsqueeze(1)
+    div_term = torch.exp(torch.arange(0, d_model, 2) * -(math.log(10000.0) / d_model))
+    pe[:, 0::2] = torch.sin(position * div_term)
+    pe[:, 1::2] = torch.cos(position * div_term)
+    if normalize:
+        pe = pe - pe.mean()
+        pe = pe / (pe.std() * 10)
+    return pe
+
+SinCosPosEncoding = PositionalEncoding
+
+def Coord2dPosEncoding(q_len, d_model, exponential=False, normalize=True, eps=1e-3, verbose=False):
+    x = .5 if exponential else 1
+    i = 0
+    for i in range(100):
+        cpe = 2 * (torch.linspace(0, 1, q_len).reshape(-1, 1) ** x) * (torch.linspace(0, 1, d_model).reshape(1, -1) ** x) - 1
+        if abs(cpe.mean()) <= eps:
+            break
+        elif cpe.mean() > eps: 
+            x += .001
+        else: 
+            x -= .001
+        i += 1
+    if normalize:
+        cpe = cpe - cpe.mean()
+        cpe = cpe / (cpe.std() * 10)
+    return cpe
+
+def Coord1dPosEncoding(q_len, exponential=False, normalize=True):
+    cpe = (2 * (torch.linspace(0, 1, q_len).reshape(-1, 1)**(.5 if exponential else 1)) - 1)
+    if normalize:
+        cpe = cpe - cpe.mean()
+        cpe = cpe / (cpe.std() * 10)
+    return cpe
+
+def positional_encoding(pe, learn_pe, q_len, d_model):
+    # Positional encoding
+    if pe is None:
+        W_pos = torch.empty((q_len, d_model)) # pe = None and learn_pe = False can be used to measure impact of pe
+        nn.init.uniform_(W_pos, -0.02, 0.02)
+        learn_pe = False
+    elif pe == 'zero':
+        W_pos = torch.empty((q_len, 1))
+        nn.init.uniform_(W_pos, -0.02, 0.02)
+    elif pe == 'zeros':
+        W_pos = torch.empty((q_len, d_model))
+        nn.init.uniform_(W_pos, -0.02, 0.02)
+    elif pe == 'normal' or pe == 'gauss':
+        W_pos = torch.zeros((q_len, 1))
+        torch.nn.init.normal_(W_pos, mean=0.0, std=0.1)
+    elif pe == 'uniform':
+        W_pos = torch.zeros((q_len, 1))
+        nn.init.uniform_(W_pos, a=0.0, b=0.1)
+    elif pe == 'lin1d': 
+        W_pos = Coord1dPosEncoding(q_len, exponential=False, normalize=True)
+    elif pe == 'exp1d': 
+        W_pos = Coord1dPosEncoding(q_len, exponential=True, normalize=True)
+    elif pe == 'lin2d':
+        W_pos = Coord2dPosEncoding(q_len, d_model, exponential=False, normalize=True)
+    elif pe == 'exp2d': 
+        W_pos = Coord2dPosEncoding(q_len, d_model, exponential=True, normalize=True)
+    elif pe == 'sincos': 
+        W_pos = PositionalEncoding(q_len, d_model, normalize=True)
+    else: 
+        raise ValueError(f"{pe} is not a valid pe (positional encoder. Available types: 'gauss'=='normal', \
+        'zeros', 'zero', uniform', 'lin1d', 'exp1d', 'lin2d', 'exp2d', 'sincos', None.)")
+    return nn.Parameter(W_pos, requires_grad=learn_pe)
+
+
+
+class PatchTST_backbone(nn.Module):
+    def __init__(self, c_in:int, context_window:int, target_window:int, patch_len:int, stride:int, max_seq_len:Optional[int]=1024, 
+                 n_layers:int=3, d_model=128, n_heads=16, d_k:Optional[int]=None, d_v:Optional[int]=None,
+                 d_ff:int=256, norm:str='BatchNorm', attn_dropout:float=0., dropout:float=0., act:str="gelu", key_padding_mask:bool='auto',
+                 padding_var:Optional[int]=None, attn_mask:Optional[Tensor]=None, res_attention:bool=True, pre_norm:bool=False, store_attn:bool=False,
+                 pe:str='zeros', learn_pe:bool=True, fc_dropout:float=0., head_dropout = 0, padding_patch = None,
+                 pretrain_head:bool=False, head_type = 'flatten', individual = False, revin = True, affine = True, subtract_last = False,
+                 verbose:bool=False, **kwargs):
+        
+        super().__init__()
+        
+        # RevIn
+        self.revin = revin
+        if self.revin: 
+            self.revin_layer = RevIN(c_in, affine=affine, subtract_last=subtract_last)
+        
+        # Patching
+        self.patch_len = patch_len
+        self.stride = stride
+        self.padding_patch = padding_patch
+        patch_num = int((context_window - patch_len)/stride + 1)
+        if padding_patch == 'end': # can be modified to general case
+            self.padding_patch_layer = nn.ReplicationPad1d((0, stride)) 
+            patch_num += 1
+        
+        # Backbone 
+        self.backbone = TSTiEncoder(c_in, patch_num=patch_num, patch_len=patch_len, max_seq_len=max_seq_len,
+                                n_layers=n_layers, d_model=d_model, n_heads=n_heads, d_k=d_k, d_v=d_v, d_ff=d_ff,
+                                attn_dropout=attn_dropout, dropout=dropout, act=act, key_padding_mask=key_padding_mask, padding_var=padding_var,
+                                attn_mask=attn_mask, res_attention=res_attention, pre_norm=pre_norm, store_attn=store_attn,
+                                pe=pe, learn_pe=learn_pe, verbose=verbose, **kwargs)
+
+        # Head
+        self.head_nf = d_model * patch_num
+        self.n_vars = c_in
+        self.pretrain_head = pretrain_head
+        self.head_type = head_type
+        self.individual = individual
+
+        if self.pretrain_head: 
+            self.head = self.create_pretrain_head(self.head_nf, c_in, fc_dropout) # custom head passed as a partial func with all its kwargs
+        elif head_type == 'flatten': 
+            self.head = Flatten_Head(self.individual, self.n_vars, self.head_nf, target_window, head_dropout=head_dropout)
+        
+    
+    def forward(self, z):                                                                   # z: [bs x nvars x seq_len]
+        # norm
+        if self.revin: 
+            z = z.permute(0,2,1)
+            z = self.revin_layer(z, 'norm')
+            z = z.permute(0,2,1)
+            
+        # do patching
+        if self.padding_patch == 'end':
+            z = self.padding_patch_layer(z)
+        z = z.unfold(dimension=-1, size=self.patch_len, step=self.stride)                   # z: [bs x nvars x patch_num x patch_len]
+        z = z.permute(0,1,3,2)                                                              # z: [bs x nvars x patch_len x patch_num]
+        
+        # model
+        z = self.backbone(z)                                                                # z: [bs x nvars x d_model x patch_num]
+        z = self.head(z)                                                                    # z: [bs x nvars x target_window] 
+        
+        # denorm
+        if self.revin: 
+            z = z.permute(0,2,1)
+            z = self.revin_layer(z, 'denorm')
+            z = z.permute(0,2,1)
+        return z
+    
+    def create_pretrain_head(self, head_nf, vars, dropout):
+        return nn.Sequential(nn.Dropout(dropout),
+                    nn.Conv1d(head_nf, vars, 1)
+                    )
+
+
+class Flatten_Head(nn.Module):
+    def __init__(self, individual, n_vars, nf, target_window, head_dropout=0):
+        super().__init__()
+        
+        self.individual = individual
+        self.n_vars = n_vars
+        
+        if self.individual:
+            self.linears = nn.ModuleList()
+            self.dropouts = nn.ModuleList()
+            self.flattens = nn.ModuleList()
+            for i in range(self.n_vars):
+                self.flattens.append(nn.Flatten(start_dim=-2))
+                self.linears.append(nn.Linear(nf, target_window))
+                self.dropouts.append(nn.Dropout(head_dropout))
+        else:
+            self.flatten = nn.Flatten(start_dim=-2)
+            self.linear = nn.Linear(nf, target_window)
+            self.dropout = nn.Dropout(head_dropout)
+            
+    def forward(self, x):                                 # x: [bs x nvars x d_model x patch_num]
+        if self.individual:
+            x_out = []
+            for i in range(self.n_vars):
+                z = self.flattens[i](x[:,i,:,:])          # z: [bs x d_model * patch_num]
+                z = self.linears[i](z)                    # z: [bs x target_window]
+                z = self.dropouts[i](z)
+                x_out.append(z)
+            x = torch.stack(x_out, dim=1)                 # x: [bs x nvars x target_window]
+        else:
+            x = self.flatten(x)
+            x = self.linear(x)
+            x = self.dropout(x)
+        return x
+        
+        
+    
+    
+class TSTiEncoder(nn.Module):  #i means channel-independent
+    def __init__(self, c_in, patch_num, patch_len, max_seq_len=1024,
+                 n_layers=3, d_model=128, n_heads=16, d_k=None, d_v=None,
+                 d_ff=256, norm='BatchNorm', attn_dropout=0., dropout=0., act="gelu", store_attn=False,
+                 key_padding_mask='auto', padding_var=None, attn_mask=None, res_attention=True, pre_norm=False,
+                 pe='zeros', learn_pe=True, verbose=False, **kwargs):
+        
+        
+        super().__init__()
+        
+        self.patch_num = patch_num
+        self.patch_len = patch_len
+        
+        # Input encoding
+        q_len = patch_num
+        self.W_P = nn.Linear(patch_len, d_model)        # Eq 1: projection of feature vectors onto a d-dim vector space
+        self.seq_len = q_len
+
+        # Positional encoding
+        self.W_pos = positional_encoding(pe, learn_pe, q_len, d_model)
+
+        # Residual dropout
+        self.dropout = nn.Dropout(dropout)
+
+        # Encoder
+        self.encoder = TSTEncoder(q_len, d_model, n_heads, d_k=d_k, d_v=d_v, d_ff=d_ff, norm=norm, attn_dropout=attn_dropout, dropout=dropout,
+                                   pre_norm=pre_norm, activation=act, res_attention=res_attention, n_layers=n_layers, store_attn=store_attn)
+
+        
+    def forward(self, x) -> Tensor:                                              # x: [bs x nvars x patch_len x patch_num]
+        
+        n_vars = x.shape[1]
+        # Input encoding
+        x = x.permute(0,1,3,2)                                                   # x: [bs x nvars x patch_num x patch_len]
+        x = self.W_P(x)                                                          # x: [bs x nvars x patch_num x d_model]
+
+        u = torch.reshape(x, (x.shape[0]*x.shape[1],x.shape[2],x.shape[3]))      # u: [bs * nvars x patch_num x d_model]
+        u = self.dropout(u + self.W_pos)                                         # u: [bs * nvars x patch_num x d_model]
+
+        # Encoder
+        z = self.encoder(u)                                                      # z: [bs * nvars x patch_num x d_model]
+        z = torch.reshape(z, (-1,n_vars,z.shape[-2],z.shape[-1]))                # z: [bs x nvars x patch_num x d_model]
+        z = z.permute(0,1,3,2)                                                   # z: [bs x nvars x d_model x patch_num]
+        
+        return z    
+            
+            
+    
+# Cell
+class TSTEncoder(nn.Module):
+    def __init__(self, q_len, d_model, n_heads, d_k=None, d_v=None, d_ff=None, 
+                        norm='BatchNorm', attn_dropout=0., dropout=0., activation='gelu',
+                        res_attention=False, n_layers=1, pre_norm=False, store_attn=False):
+        super().__init__()
+
+        self.layers = nn.ModuleList([TSTEncoderLayer(q_len, d_model, n_heads=n_heads, d_k=d_k, d_v=d_v, d_ff=d_ff, norm=norm,
+                                                      attn_dropout=attn_dropout, dropout=dropout,
+                                                      activation=activation, res_attention=res_attention,
+                                                      pre_norm=pre_norm, store_attn=store_attn) for _ in range(n_layers)])
+        self.res_attention = res_attention
+
+    def forward(self, src:Tensor, key_padding_mask:Optional[Tensor]=None, attn_mask:Optional[Tensor]=None):
+        output = src
+        scores = None
+        if self.res_attention:
+            for mod in self.layers: 
+                output, scores = mod(output, prev=scores, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+            return output
+        else:
+            for mod in self.layers:
+                output = mod(output, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+            return output
+
+
+
+class TSTEncoderLayer(nn.Module):
+    def __init__(self, q_len, d_model, n_heads, d_k=None, d_v=None, d_ff=256, store_attn=False,
+                 norm='BatchNorm', attn_dropout=0, dropout=0., bias=True, activation="gelu", res_attention=False, pre_norm=False):
+        super().__init__()
+        assert not d_model%n_heads, f"d_model ({d_model}) must be divisible by n_heads ({n_heads})"
+        d_k = d_model // n_heads if d_k is None else d_k
+        d_v = d_model // n_heads if d_v is None else d_v
+
+        # Multi-Head attention
+        self.res_attention = res_attention
+        self.self_attn = _MultiheadAttention(d_model, n_heads, d_k, d_v, attn_dropout=attn_dropout, proj_dropout=dropout, res_attention=res_attention)
+
+        # Add & Norm
+        self.dropout_attn = nn.Dropout(dropout)
+        if "batch" in norm.lower():
+            self.norm_attn = nn.Sequential(Transpose(1,2), nn.BatchNorm1d(d_model), Transpose(1,2))
+        else:
+            self.norm_attn = nn.LayerNorm(d_model)
+
+        # Position-wise Feed-Forward
+        self.ff = nn.Sequential(nn.Linear(d_model, d_ff, bias=bias),
+                                activation,
+                                nn.Dropout(dropout),
+                                nn.Linear(d_ff, d_model, bias=bias))
+
+        # Add & Norm
+        self.dropout_ffn = nn.Dropout(dropout)
+        if "batch" in norm.lower():
+            self.norm_ffn = nn.Sequential(Transpose(1,2), nn.BatchNorm1d(d_model), Transpose(1,2))
+        else:
+            self.norm_ffn = nn.LayerNorm(d_model)
+
+        self.pre_norm = pre_norm
+        self.store_attn = store_attn
+
+
+    def forward(self, src:Tensor, prev:Optional[Tensor]=None, key_padding_mask:Optional[Tensor]=None, attn_mask:Optional[Tensor]=None) -> Tensor:
+
+        # Multi-Head attention sublayer
+        if self.pre_norm:
+            src = self.norm_attn(src)
+        ## Multi-Head attention
+        if self.res_attention:
+            src2, attn, scores = self.self_attn(src, src, src, prev, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+        else:
+            src2, attn = self.self_attn(src, src, src, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+        if self.store_attn:
+            self.attn = attn
+        ## Add & Norm
+        src = src + self.dropout_attn(src2) # Add: residual connection with residual dropout
+        if not self.pre_norm:
+            src = self.norm_attn(src)
+
+        # Feed-forward sublayer
+        if self.pre_norm:
+            src = self.norm_ffn(src)
+        ## Position-wise Feed-Forward
+        src2 = self.ff(src)
+        ## Add & Norm
+        src = src + self.dropout_ffn(src2) # Add: residual connection with residual dropout
+        if not self.pre_norm:
+            src = self.norm_ffn(src)
+
+        if self.res_attention:
+            return src, scores
+        else:
+            return src
+
+
+
+
+class _MultiheadAttention(nn.Module):
+    def __init__(self, d_model, n_heads, d_k=None, d_v=None, res_attention=False, attn_dropout=0., proj_dropout=0., qkv_bias=True, lsa=False):
+        """Multi Head Attention Layer
+        Input shape:
+            Q:       [batch_size (bs) x max_q_len x d_model]
+            K, V:    [batch_size (bs) x q_len x d_model]
+            mask:    [q_len x q_len]
+        """
+        super().__init__()
+        d_k = d_model // n_heads if d_k is None else d_k
+        d_v = d_model // n_heads if d_v is None else d_v
+
+        self.n_heads, self.d_k, self.d_v = n_heads, d_k, d_v
+
+        self.W_Q = nn.Linear(d_model, d_k * n_heads, bias=qkv_bias)
+        self.W_K = nn.Linear(d_model, d_k * n_heads, bias=qkv_bias)
+        self.W_V = nn.Linear(d_model, d_v * n_heads, bias=qkv_bias)
+
+        # Scaled Dot-Product Attention (multiple heads)
+        self.res_attention = res_attention
+        self.sdp_attn = _ScaledDotProductAttention(d_model, n_heads, attn_dropout=attn_dropout, res_attention=self.res_attention, lsa=lsa)
+
+        # Poject output
+        self.to_out = nn.Sequential(nn.Linear(n_heads * d_v, d_model), nn.Dropout(proj_dropout))
+
+
+    def forward(self, Q:Tensor, K:Optional[Tensor]=None, V:Optional[Tensor]=None, prev:Optional[Tensor]=None,
+                key_padding_mask:Optional[Tensor]=None, attn_mask:Optional[Tensor]=None):
+
+        bs = Q.size(0)
+        if K is None: 
+            K = Q
+        if V is None: 
+            V = Q
+
+        # Linear (+ split in multiple heads)
+        q_s = self.W_Q(Q).view(bs, -1, self.n_heads, self.d_k).transpose(1,2)       # q_s    : [bs x n_heads x max_q_len x d_k]
+        k_s = self.W_K(K).view(bs, -1, self.n_heads, self.d_k).permute(0,2,3,1)     # k_s    : [bs x n_heads x d_k x q_len] - transpose(1,2) + transpose(2,3)
+        v_s = self.W_V(V).view(bs, -1, self.n_heads, self.d_v).transpose(1,2)       # v_s    : [bs x n_heads x q_len x d_v]
+
+        # Apply Scaled Dot-Product Attention (multiple heads)
+        if self.res_attention:
+            output, attn_weights, attn_scores = self.sdp_attn(q_s, k_s, v_s, prev=prev, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+        else:
+            output, attn_weights = self.sdp_attn(q_s, k_s, v_s, key_padding_mask=key_padding_mask, attn_mask=attn_mask)
+        # output: [bs x n_heads x q_len x d_v], attn: [bs x n_heads x q_len x q_len], scores: [bs x n_heads x max_q_len x q_len]
+
+        # back to the original inputs dimensions
+        output = output.transpose(1, 2).contiguous().view(bs, -1, self.n_heads * self.d_v) # output: [bs x q_len x n_heads * d_v]
+        output = self.to_out(output)
+
+        if self.res_attention: 
+            return output, attn_weights, attn_scores
+        else: 
+            return output, attn_weights
+
+
+class _ScaledDotProductAttention(nn.Module):
+    r"""Scaled Dot-Product Attention module (Attention is all you need by Vaswani et al., 2017) with optional residual attention from previous layer
+    (Realformer: Transformer likes residual attention by He et al, 2020) and locality self sttention (Vision Transformer for Small-Size Datasets
+    by Lee et al, 2021)"""
+
+    def __init__(self, d_model, n_heads, attn_dropout=0., res_attention=False, lsa=False):
+        super().__init__()
+        self.attn_dropout = nn.Dropout(attn_dropout)
+        self.res_attention = res_attention
+        head_dim = d_model // n_heads
+        self.scale = nn.Parameter(torch.tensor(head_dim ** -0.5), requires_grad=lsa)
+        self.lsa = lsa
+
+    def forward(self, q:Tensor, k:Tensor, v:Tensor, prev:Optional[Tensor]=None, key_padding_mask:Optional[Tensor]=None, attn_mask:Optional[Tensor]=None):
+        '''
+        Input shape:
+            q               : [bs x n_heads x max_q_len x d_k]
+            k               : [bs x n_heads x d_k x seq_len]
+            v               : [bs x n_heads x seq_len x d_v]
+            prev            : [bs x n_heads x q_len x seq_len]
+            key_padding_mask: [bs x seq_len]
+            attn_mask       : [1 x seq_len x seq_len]
+        Output shape:
+            output:  [bs x n_heads x q_len x d_v]
+            attn   : [bs x n_heads x q_len x seq_len]
+            scores : [bs x n_heads x q_len x seq_len]
+        '''
+
+        # Scaled MatMul (q, k) - similarity scores for all pairs of positions in an input sequence
+        attn_scores = torch.matmul(q, k) * self.scale      # attn_scores : [bs x n_heads x max_q_len x q_len]
+
+        # Add pre-softmax attention scores from the previous layer (optional)
+        if prev is not None: 
+            attn_scores = attn_scores + prev
+
+        # Attention mask (optional)
+        if attn_mask is not None:                                     # attn_mask with shape [q_len x seq_len] - only used when q_len == seq_len
+            if attn_mask.dtype == torch.bool:
+                attn_scores.masked_fill_(attn_mask, -np.inf)
+            else:
+                attn_scores += attn_mask
+
+        # Key padding mask (optional)
+        if key_padding_mask is not None:                              # mask with shape [bs x q_len] (only when max_w_len == q_len)
+            attn_scores.masked_fill_(key_padding_mask.unsqueeze(1).unsqueeze(2), -np.inf)
+
+        # normalize the attention weights
+        attn_weights = F.softmax(attn_scores, dim=-1)                 # attn_weights   : [bs x n_heads x max_q_len x q_len]
+        attn_weights = self.attn_dropout(attn_weights)
+
+        # compute the new values given the attention weights
+        output = torch.matmul(attn_weights, v)                        # output: [bs x n_heads x max_q_len x d_v]
+
+        if self.res_attention: 
+            return output, attn_weights, attn_scores
+        else: 
+            return output, attn_weights
+        
+        
+
+class RevIN(nn.Module):
+    def __init__(self, num_features: int, eps=1e-5, affine=True, subtract_last=False):
+        """
+        :param num_features: the number of features or channels
+        :param eps: a value added for numerical stability
+        :param affine: if True, RevIN has learnable affine parameters
+        """
+        super(RevIN, self).__init__()
+        self.num_features = num_features
+        self.eps = eps
+        self.affine = affine
+        self.subtract_last = subtract_last
+        if self.affine:
+            self._init_params()
+
+    def forward(self, x, mode:str):
+        if mode == 'norm':
+            self._get_statistics(x)
+            x = self._normalize(x)
+        elif mode == 'denorm':
+            x = self._denormalize(x)
+        else: 
+            raise NotImplementedError
+        return x
+
+    def _init_params(self):
+        # initialize RevIN params: (C,)
+        self.affine_weight = nn.Parameter(torch.ones(self.num_features))
+        self.affine_bias = nn.Parameter(torch.zeros(self.num_features))
+
+    def _get_statistics(self, x):
+        dim2reduce = tuple(range(1, x.ndim-1))
+        if self.subtract_last:
+            self.last = x[:,-1,:].unsqueeze(1)
+        else:
+            self.mean = torch.mean(x, dim=dim2reduce, keepdim=True).detach()
+        self.stdev = torch.sqrt(torch.var(x, dim=dim2reduce, keepdim=True, unbiased=False) + self.eps).detach()
+
+    def _normalize(self, x):
+        if self.subtract_last:
+            x = x - self.last
+        else:
+            x = x - self.mean
+        x = x / self.stdev
+        if self.affine:
+            x = x * self.affine_weight
+            x = x + self.affine_bias
+        return x
+
+    def _denormalize(self, x):
+        if self.affine:
+            x = x - self.affine_bias
+            x = x / (self.affine_weight + self.eps*self.eps)
+        x = x * self.stdev
+        if self.subtract_last:
+            x = x + self.last
+        else:
+            x = x + self.mean
+        return x